From dd67b6769a051b286d101a7264a2fcefc6ddb5ea Mon Sep 17 00:00:00 2001 From: Morten Hjorth-Jensen Date: Mon, 14 Oct 2024 06:03:15 +0200 Subject: [PATCH] update --- doc/LectureNotes/figures/simplenn3.png | Bin 525739 -> 3751 bytes doc/pub/week42/html/._week42-bs000.html | 91 +- doc/pub/week42/html/._week42-bs001.html | 91 +- doc/pub/week42/html/._week42-bs002.html | 91 +- doc/pub/week42/html/._week42-bs003.html | 91 +- doc/pub/week42/html/._week42-bs004.html | 91 +- doc/pub/week42/html/._week42-bs005.html | 91 +- doc/pub/week42/html/._week42-bs006.html | 91 +- doc/pub/week42/html/._week42-bs007.html | 91 +- doc/pub/week42/html/._week42-bs008.html | 91 +- doc/pub/week42/html/._week42-bs009.html | 91 +- doc/pub/week42/html/._week42-bs010.html | 91 +- doc/pub/week42/html/._week42-bs011.html | 91 +- doc/pub/week42/html/._week42-bs012.html | 91 +- doc/pub/week42/html/._week42-bs013.html | 91 +- doc/pub/week42/html/._week42-bs014.html | 91 +- doc/pub/week42/html/._week42-bs015.html | 91 +- doc/pub/week42/html/._week42-bs016.html | 91 +- doc/pub/week42/html/._week42-bs017.html | 93 +- doc/pub/week42/html/._week42-bs018.html | 97 +- doc/pub/week42/html/._week42-bs019.html | 95 +- doc/pub/week42/html/._week42-bs020.html | 93 +- doc/pub/week42/html/._week42-bs021.html | 91 +- doc/pub/week42/html/._week42-bs022.html | 99 +- doc/pub/week42/html/._week42-bs023.html | 97 +- doc/pub/week42/html/._week42-bs024.html | 97 +- doc/pub/week42/html/._week42-bs025.html | 95 +- doc/pub/week42/html/._week42-bs026.html | 91 +- doc/pub/week42/html/._week42-bs027.html | 91 +- doc/pub/week42/html/._week42-bs028.html | 91 +- doc/pub/week42/html/._week42-bs029.html | 91 +- doc/pub/week42/html/._week42-bs030.html | 91 +- doc/pub/week42/html/._week42-bs031.html | 91 +- doc/pub/week42/html/._week42-bs032.html | 91 +- doc/pub/week42/html/._week42-bs033.html | 91 +- doc/pub/week42/html/._week42-bs034.html | 91 +- doc/pub/week42/html/._week42-bs035.html | 91 +- doc/pub/week42/html/._week42-bs036.html | 91 +- doc/pub/week42/html/._week42-bs037.html | 91 +- doc/pub/week42/html/._week42-bs038.html | 91 +- doc/pub/week42/html/._week42-bs039.html | 91 +- doc/pub/week42/html/._week42-bs040.html | 91 +- doc/pub/week42/html/._week42-bs041.html | 120 +- doc/pub/week42/html/._week42-bs042.html | 113 +- doc/pub/week42/html/._week42-bs043.html | 106 +- doc/pub/week42/html/._week42-bs044.html | 103 +- doc/pub/week42/html/._week42-bs045.html | 119 +- doc/pub/week42/html/._week42-bs046.html | 122 +- doc/pub/week42/html/._week42-bs047.html | 203 +- doc/pub/week42/html/._week42-bs048.html | 210 +- doc/pub/week42/html/._week42-bs049.html | 119 +- doc/pub/week42/html/._week42-bs050.html | 114 +- doc/pub/week42/html/._week42-bs051.html | 115 +- doc/pub/week42/html/._week42-bs052.html | 116 +- doc/pub/week42/html/._week42-bs053.html | 119 +- doc/pub/week42/html/._week42-bs054.html | 120 +- doc/pub/week42/html/._week42-bs055.html | 113 +- doc/pub/week42/html/._week42-bs056.html | 110 +- doc/pub/week42/html/._week42-bs057.html | 111 +- doc/pub/week42/html/._week42-bs058.html | 113 +- doc/pub/week42/html/._week42-bs059.html | 121 +- doc/pub/week42/html/._week42-bs060.html | 121 +- doc/pub/week42/html/._week42-bs061.html | 115 +- doc/pub/week42/html/._week42-bs062.html | 109 +- doc/pub/week42/html/._week42-bs063.html | 119 +- doc/pub/week42/html/._week42-bs064.html | 128 +- doc/pub/week42/html/._week42-bs065.html | 119 +- doc/pub/week42/html/._week42-bs066.html | 106 +- doc/pub/week42/html/._week42-bs067.html | 98 +- doc/pub/week42/html/._week42-bs068.html | 97 +- doc/pub/week42/html/._week42-bs069.html | 163 +- doc/pub/week42/html/._week42-bs070.html | 91 +- doc/pub/week42/html/._week42-bs071.html | 91 +- doc/pub/week42/html/._week42-bs072.html | 91 +- doc/pub/week42/html/._week42-bs073.html | 91 +- doc/pub/week42/html/._week42-bs074.html | 91 +- doc/pub/week42/html/._week42-bs075.html | 91 +- doc/pub/week42/html/._week42-bs076.html | 91 +- doc/pub/week42/html/._week42-bs077.html | 91 +- doc/pub/week42/html/._week42-bs078.html | 91 +- doc/pub/week42/html/._week42-bs079.html | 91 +- doc/pub/week42/html/._week42-bs080.html | 91 +- doc/pub/week42/html/._week42-bs081.html | 91 +- doc/pub/week42/html/._week42-bs082.html | 91 +- doc/pub/week42/html/._week42-bs083.html | 91 +- doc/pub/week42/html/._week42-bs084.html | 91 +- doc/pub/week42/html/._week42-bs085.html | 91 +- doc/pub/week42/html/._week42-bs086.html | 91 +- doc/pub/week42/html/._week42-bs087.html | 91 +- doc/pub/week42/html/._week42-bs088.html | 91 +- doc/pub/week42/html/._week42-bs089.html | 91 +- doc/pub/week42/html/._week42-bs090.html | 91 +- doc/pub/week42/html/._week42-bs091.html | 91 +- doc/pub/week42/html/._week42-bs092.html | 91 +- doc/pub/week42/html/figures/simplenn3.png | Bin 525739 -> 3751 bytes doc/pub/week42/html/week42-bs.html | 91 +- doc/pub/week42/html/week42-reveal.html | 242 +- doc/pub/week42/html/week42-solarized.html | 258 +- doc/pub/week42/html/week42.html | 258 +- doc/pub/week42/ipynb/ipynb-week42-src.tar.gz | Bin 978624 -> 487084 bytes doc/pub/week42/ipynb/week42.ipynb | 2490 ++++--- doc/src/week42/._week42-bs000.html | 529 ++ doc/src/week42/._week42-bs001.html | 530 ++ doc/src/week42/._week42-bs002.html | 517 ++ doc/src/week42/._week42-bs003.html | 517 ++ doc/src/week42/._week42-bs004.html | 518 ++ doc/src/week42/._week42-bs005.html | 517 ++ doc/src/week42/._week42-bs006.html | 513 ++ doc/src/week42/._week42-bs007.html | 532 ++ doc/src/week42/._week42-bs008.html | 517 ++ doc/src/week42/._week42-bs009.html | 538 ++ doc/src/week42/._week42-bs010.html | 535 ++ doc/src/week42/._week42-bs011.html | 519 ++ doc/src/week42/._week42-bs012.html | 533 ++ doc/src/week42/._week42-bs013.html | 526 ++ doc/src/week42/._week42-bs014.html | 530 ++ doc/src/week42/._week42-bs015.html | 607 ++ doc/src/week42/._week42-bs016.html | 531 ++ doc/src/week42/._week42-bs017.html | 519 ++ doc/src/week42/._week42-bs018.html | 526 ++ doc/src/week42/._week42-bs019.html | 526 ++ doc/src/week42/._week42-bs020.html | 524 ++ doc/src/week42/._week42-bs021.html | 535 ++ doc/src/week42/._week42-bs022.html | 530 ++ doc/src/week42/._week42-bs023.html | 528 ++ doc/src/week42/._week42-bs024.html | 529 ++ doc/src/week42/._week42-bs025.html | 525 ++ doc/src/week42/._week42-bs026.html | 537 ++ doc/src/week42/._week42-bs027.html | 531 ++ doc/src/week42/._week42-bs028.html | 519 ++ doc/src/week42/._week42-bs029.html | 536 ++ doc/src/week42/._week42-bs030.html | 525 ++ doc/src/week42/._week42-bs031.html | 519 ++ doc/src/week42/._week42-bs032.html | 529 ++ doc/src/week42/._week42-bs033.html | 532 ++ doc/src/week42/._week42-bs034.html | 529 ++ doc/src/week42/._week42-bs035.html | 522 ++ doc/src/week42/._week42-bs036.html | 532 ++ doc/src/week42/._week42-bs037.html | 526 ++ doc/src/week42/._week42-bs038.html | 540 ++ doc/src/week42/._week42-bs039.html | 520 ++ doc/src/week42/._week42-bs040.html | 532 ++ doc/src/week42/._week42-bs041.html | 529 ++ doc/src/week42/._week42-bs042.html | 528 ++ doc/src/week42/._week42-bs043.html | 524 ++ doc/src/week42/._week42-bs044.html | 529 ++ doc/src/week42/._week42-bs045.html | 529 ++ doc/src/week42/._week42-bs046.html | 525 ++ doc/src/week42/._week42-bs047.html | 534 ++ doc/src/week42/._week42-bs048.html | 614 ++ doc/src/week42/._week42-bs049.html | 527 ++ doc/src/week42/._week42-bs050.html | 522 ++ doc/src/week42/._week42-bs051.html | 528 ++ doc/src/week42/._week42-bs052.html | 523 ++ doc/src/week42/._week42-bs053.html | 533 ++ doc/src/week42/._week42-bs054.html | 524 ++ doc/src/week42/._week42-bs055.html | 526 ++ doc/src/week42/._week42-bs056.html | 523 ++ doc/src/week42/._week42-bs057.html | 527 ++ doc/src/week42/._week42-bs058.html | 527 ++ doc/src/week42/._week42-bs059.html | 535 ++ doc/src/week42/._week42-bs060.html | 531 ++ doc/src/week42/._week42-bs061.html | 531 ++ doc/src/week42/._week42-bs062.html | 525 ++ doc/src/week42/._week42-bs063.html | 525 ++ doc/src/week42/._week42-bs064.html | 529 ++ doc/src/week42/._week42-bs065.html | 534 ++ doc/src/week42/._week42-bs066.html | 522 ++ doc/src/week42/._week42-bs067.html | 519 ++ doc/src/week42/._week42-bs068.html | 516 ++ doc/src/week42/._week42-bs069.html | 518 ++ doc/src/week42/._week42-bs070.html | 547 ++ doc/src/week42/._week42-bs071.html | 562 ++ doc/src/week42/._week42-bs072.html | 556 ++ doc/src/week42/._week42-bs073.html | 530 ++ doc/src/week42/._week42-bs074.html | 523 ++ doc/src/week42/._week42-bs075.html | 622 ++ doc/src/week42/._week42-bs076.html | 574 ++ doc/src/week42/._week42-bs077.html | 552 ++ doc/src/week42/._week42-bs078.html | 551 ++ doc/src/week42/._week42-bs079.html | 562 ++ doc/src/week42/._week42-bs080.html | 533 ++ doc/src/week42/._week42-bs081.html | 600 ++ doc/src/week42/._week42-bs082.html | 538 ++ doc/src/week42/._week42-bs083.html | 545 ++ doc/src/week42/._week42-bs084.html | 540 ++ doc/src/week42/._week42-bs085.html | 637 ++ doc/src/week42/._week42-bs086.html | 527 ++ doc/src/week42/._week42-bs087.html | 639 ++ doc/src/week42/._week42-bs088.html | 561 ++ doc/src/week42/._week42-bs089.html | 560 ++ doc/src/week42/._week42-bs090.html | 567 ++ doc/src/week42/._week42-bs091.html | 563 ++ doc/src/week42/._week42-bs092.html | 565 ++ doc/src/week42/._week42-bs093.html | 518 ++ doc/src/week42/._week42-bs094.html | 610 ++ doc/src/week42/._week42-bs095.html | 537 ++ doc/src/week42/._week42-bs096.html | 729 ++ doc/src/week42/._week42-bs097.html | 692 ++ doc/src/week42/._week42-bs098.html | 1809 +++++ doc/src/week42/._week42-bs099.html | 538 ++ doc/src/week42/figures/simplenn3.png | Bin 0 -> 3751 bytes doc/src/week42/ipynb-week42-src.tar.gz | Bin 0 -> 487084 bytes doc/src/week42/reveal.js/.gitignore | 8 + doc/src/week42/reveal.js/.travis.yml | 5 + doc/src/week42/reveal.js/CONTRIBUTING.md | 23 + doc/src/week42/reveal.js/Gruntfile.js | 140 + doc/src/week42/reveal.js/LICENSE | 19 + doc/src/week42/reveal.js/README.md | 1052 +++ doc/src/week42/reveal.js/bower.json | 27 + .../reveal.js/css/images/cbc_footer.png | Bin 0 -> 10008 bytes .../reveal.js/css/images/cbc_symbol.png | Bin 0 -> 2946 bytes .../reveal.js/css/images/simula_footer.png | Bin 0 -> 2513 bytes .../reveal.js/css/images/simula_logo.png | Bin 0 -> 2138 bytes .../reveal.js/css/images/simula_symbol.png | Bin 0 -> 2138 bytes .../reveal.js/css/images/uio_footer.png | Bin 0 -> 18189 bytes .../reveal.js/css/images/uio_symbol.png | Bin 0 -> 11352 bytes doc/src/week42/reveal.js/css/print/paper.css | 202 + doc/src/week42/reveal.js/css/print/pdf.css | 157 + doc/src/week42/reveal.js/css/reveal.css | 1886 +++++ doc/src/week42/reveal.js/css/reveal.scss | 1319 ++++ doc/src/week42/reveal.js/css/theme/README.md | 23 + doc/src/week42/reveal.js/css/theme/beige.css | 154 + .../week42/reveal.js/css/theme/beigesmall.css | 155 + doc/src/week42/reveal.js/css/theme/black.css | 273 + doc/src/week42/reveal.js/css/theme/blood.css | 180 + doc/src/week42/reveal.js/css/theme/cbc.css | 144 + .../week42/reveal.js/css/theme/darkgray.css | 153 + .../week42/reveal.js/css/theme/default.css | 153 + doc/src/week42/reveal.js/css/theme/league.css | 279 + doc/src/week42/reveal.js/css/theme/moon.css | 153 + doc/src/week42/reveal.js/css/theme/night.css | 141 + doc/src/week42/reveal.js/css/theme/serif.css | 143 + doc/src/week42/reveal.js/css/theme/simple.css | 144 + doc/src/week42/reveal.js/css/theme/simula.css | 144 + doc/src/week42/reveal.js/css/theme/sky.css | 150 + .../week42/reveal.js/css/theme/solarized.css | 153 + .../reveal.js/css/theme/source/beige.scss | 50 + .../css/theme/source/beigesmall.scss | 51 + .../reveal.js/css/theme/source/black.scss | 49 + .../reveal.js/css/theme/source/blood.scss | 91 + .../reveal.js/css/theme/source/cbc.scss | 39 + .../reveal.js/css/theme/source/darkgray.scss | 42 + .../reveal.js/css/theme/source/default.scss | 42 + .../reveal.js/css/theme/source/league.scss | 34 + .../reveal.js/css/theme/source/moon.scss | 68 + .../reveal.js/css/theme/source/night.scss | 35 + .../reveal.js/css/theme/source/serif.scss | 35 + .../reveal.js/css/theme/source/simple.scss | 38 + .../reveal.js/css/theme/source/simula.scss | 39 + .../reveal.js/css/theme/source/sky.scss | 46 + .../reveal.js/css/theme/source/solarized.scss | 74 + .../reveal.js/css/theme/source/white.scss | 49 + doc/src/week42/reveal.js/css/theme/white.css | 273 + doc/src/week42/reveal.js/index.html | 411 ++ doc/src/week42/reveal.js/js/reveal.js | 4508 ++++++++++++ doc/src/week42/reveal.js/lib/css/zenburn.css | 117 + .../reveal.js/lib/font/league-gothic/LICENSE | 2 + .../lib/font/league-gothic/league-gothic.css | 10 + .../lib/font/league-gothic/league-gothic.eot | Bin 0 -> 25696 bytes .../lib/font/league-gothic/league-gothic.ttf | Bin 0 -> 64256 bytes .../lib/font/league-gothic/league-gothic.woff | Bin 0 -> 30764 bytes .../lib/font/source-sans-pro/LICENSE | 45 + .../source-sans-pro-italic.eot | Bin 0 -> 75720 bytes .../source-sans-pro-italic.ttf | Bin 0 -> 238084 bytes .../source-sans-pro-italic.woff | Bin 0 -> 98556 bytes .../source-sans-pro-regular.eot | Bin 0 -> 88070 bytes .../source-sans-pro-regular.ttf | Bin 0 -> 288008 bytes .../source-sans-pro-regular.woff | Bin 0 -> 114324 bytes .../source-sans-pro-semibold.eot | Bin 0 -> 89897 bytes .../source-sans-pro-semibold.ttf | Bin 0 -> 284640 bytes .../source-sans-pro-semibold.woff | Bin 0 -> 115648 bytes .../source-sans-pro-semibolditalic.eot | Bin 0 -> 75706 bytes .../source-sans-pro-semibolditalic.ttf | Bin 0 -> 240944 bytes .../source-sans-pro-semibolditalic.woff | Bin 0 -> 98816 bytes .../font/source-sans-pro/source-sans-pro.css | 39 + doc/src/week42/reveal.js/lib/js/classList.js | 2 + doc/src/week42/reveal.js/lib/js/head.min.js | 8 + doc/src/week42/reveal.js/lib/js/html5shiv.js | 7 + doc/src/week42/reveal.js/package.json | 45 + .../reveal.js/plugin/highlight/highlight.js | 30 + doc/src/week42/reveal.js/plugin/leap/leap.js | 159 + .../reveal.js/plugin/markdown/example.html | 129 + .../reveal.js/plugin/markdown/example.md | 31 + .../reveal.js/plugin/markdown/markdown.js | 393 ++ .../reveal.js/plugin/markdown/marked.js | 6 + doc/src/week42/reveal.js/plugin/math/math.js | 64 + .../reveal.js/plugin/multiplex/client.js | 13 + .../reveal.js/plugin/multiplex/index.js | 56 + .../reveal.js/plugin/multiplex/master.js | 51 + .../reveal.js/plugin/notes-server/client.js | 60 + .../reveal.js/plugin/notes-server/index.js | 66 + .../reveal.js/plugin/notes-server/notes.html | 396 ++ .../week42/reveal.js/plugin/notes/notes.html | 406 ++ .../week42/reveal.js/plugin/notes/notes.js | 122 + .../reveal.js/plugin/print-pdf/print-pdf.js | 48 + .../reveal.js/plugin/remotes/remotes.js | 39 + .../week42/reveal.js/plugin/search/search.js | 196 + .../week42/reveal.js/plugin/zoom-js/zoom.js | 278 + .../reveal.js/test/examples/assets/image1.png | Bin 0 -> 21991 bytes .../reveal.js/test/examples/assets/image2.png | Bin 0 -> 10237 bytes .../reveal.js/test/examples/barebones.html | 41 + .../test/examples/embedded-media.html | 49 + .../week42/reveal.js/test/examples/math.html | 185 + .../test/examples/slide-backgrounds.html | 144 + .../test/examples/slide-transitions.html | 101 + .../week42/reveal.js/test/qunit-1.12.0.css | 244 + doc/src/week42/reveal.js/test/qunit-1.12.0.js | 2212 ++++++ .../test-markdown-element-attributes.html | 134 + .../test/test-markdown-element-attributes.js | 46 + .../test/test-markdown-slide-attributes.html | 128 + .../test/test-markdown-slide-attributes.js | 47 + .../week42/reveal.js/test/test-markdown.html | 52 + .../week42/reveal.js/test/test-markdown.js | 15 + doc/src/week42/reveal.js/test/test-pdf.html | 83 + doc/src/week42/reveal.js/test/test-pdf.js | 15 + doc/src/week42/reveal.js/test/test.html | 85 + doc/src/week42/reveal.js/test/test.js | 589 ++ doc/src/week42/week42-bs.html | 529 ++ doc/src/week42/week42-reveal.html | 5174 ++++++++++++++ doc/src/week42/week42-solarized.html | 4901 +++++++++++++ doc/src/week42/week42.do.txt | 211 +- doc/src/week42/week42.html | 4978 +++++++++++++ doc/src/week42/week42.ipynb | 6133 +++++++++++++++++ 324 files changed, 104185 insertions(+), 6250 deletions(-) create mode 100644 doc/src/week42/._week42-bs000.html create mode 100644 doc/src/week42/._week42-bs001.html create mode 100644 doc/src/week42/._week42-bs002.html create mode 100644 doc/src/week42/._week42-bs003.html create mode 100644 doc/src/week42/._week42-bs004.html create mode 100644 doc/src/week42/._week42-bs005.html create mode 100644 doc/src/week42/._week42-bs006.html create mode 100644 doc/src/week42/._week42-bs007.html create mode 100644 doc/src/week42/._week42-bs008.html create mode 100644 doc/src/week42/._week42-bs009.html create mode 100644 doc/src/week42/._week42-bs010.html create mode 100644 doc/src/week42/._week42-bs011.html create mode 100644 doc/src/week42/._week42-bs012.html create mode 100644 doc/src/week42/._week42-bs013.html create mode 100644 doc/src/week42/._week42-bs014.html create mode 100644 doc/src/week42/._week42-bs015.html create mode 100644 doc/src/week42/._week42-bs016.html create mode 100644 doc/src/week42/._week42-bs017.html create mode 100644 doc/src/week42/._week42-bs018.html create mode 100644 doc/src/week42/._week42-bs019.html create mode 100644 doc/src/week42/._week42-bs020.html create mode 100644 doc/src/week42/._week42-bs021.html create mode 100644 doc/src/week42/._week42-bs022.html create mode 100644 doc/src/week42/._week42-bs023.html create mode 100644 doc/src/week42/._week42-bs024.html create mode 100644 doc/src/week42/._week42-bs025.html create mode 100644 doc/src/week42/._week42-bs026.html create mode 100644 doc/src/week42/._week42-bs027.html create mode 100644 doc/src/week42/._week42-bs028.html create mode 100644 doc/src/week42/._week42-bs029.html create mode 100644 doc/src/week42/._week42-bs030.html create mode 100644 doc/src/week42/._week42-bs031.html create mode 100644 doc/src/week42/._week42-bs032.html create mode 100644 doc/src/week42/._week42-bs033.html create mode 100644 doc/src/week42/._week42-bs034.html create mode 100644 doc/src/week42/._week42-bs035.html create mode 100644 doc/src/week42/._week42-bs036.html create mode 100644 doc/src/week42/._week42-bs037.html create mode 100644 doc/src/week42/._week42-bs038.html create mode 100644 doc/src/week42/._week42-bs039.html create mode 100644 doc/src/week42/._week42-bs040.html create mode 100644 doc/src/week42/._week42-bs041.html create mode 100644 doc/src/week42/._week42-bs042.html create mode 100644 doc/src/week42/._week42-bs043.html create mode 100644 doc/src/week42/._week42-bs044.html create mode 100644 doc/src/week42/._week42-bs045.html create mode 100644 doc/src/week42/._week42-bs046.html create mode 100644 doc/src/week42/._week42-bs047.html create mode 100644 doc/src/week42/._week42-bs048.html create mode 100644 doc/src/week42/._week42-bs049.html create mode 100644 doc/src/week42/._week42-bs050.html create mode 100644 doc/src/week42/._week42-bs051.html create mode 100644 doc/src/week42/._week42-bs052.html create mode 100644 doc/src/week42/._week42-bs053.html create mode 100644 doc/src/week42/._week42-bs054.html create mode 100644 doc/src/week42/._week42-bs055.html create mode 100644 doc/src/week42/._week42-bs056.html create mode 100644 doc/src/week42/._week42-bs057.html create mode 100644 doc/src/week42/._week42-bs058.html create mode 100644 doc/src/week42/._week42-bs059.html create mode 100644 doc/src/week42/._week42-bs060.html create mode 100644 doc/src/week42/._week42-bs061.html create mode 100644 doc/src/week42/._week42-bs062.html create mode 100644 doc/src/week42/._week42-bs063.html create mode 100644 doc/src/week42/._week42-bs064.html create mode 100644 doc/src/week42/._week42-bs065.html create mode 100644 doc/src/week42/._week42-bs066.html create mode 100644 doc/src/week42/._week42-bs067.html create mode 100644 doc/src/week42/._week42-bs068.html create mode 100644 doc/src/week42/._week42-bs069.html create mode 100644 doc/src/week42/._week42-bs070.html create mode 100644 doc/src/week42/._week42-bs071.html create mode 100644 doc/src/week42/._week42-bs072.html create mode 100644 doc/src/week42/._week42-bs073.html create mode 100644 doc/src/week42/._week42-bs074.html create mode 100644 doc/src/week42/._week42-bs075.html create mode 100644 doc/src/week42/._week42-bs076.html create mode 100644 doc/src/week42/._week42-bs077.html create mode 100644 doc/src/week42/._week42-bs078.html create mode 100644 doc/src/week42/._week42-bs079.html create mode 100644 doc/src/week42/._week42-bs080.html create mode 100644 doc/src/week42/._week42-bs081.html create mode 100644 doc/src/week42/._week42-bs082.html create mode 100644 doc/src/week42/._week42-bs083.html create mode 100644 doc/src/week42/._week42-bs084.html create mode 100644 doc/src/week42/._week42-bs085.html create mode 100644 doc/src/week42/._week42-bs086.html create mode 100644 doc/src/week42/._week42-bs087.html create mode 100644 doc/src/week42/._week42-bs088.html create mode 100644 doc/src/week42/._week42-bs089.html create mode 100644 doc/src/week42/._week42-bs090.html create mode 100644 doc/src/week42/._week42-bs091.html create mode 100644 doc/src/week42/._week42-bs092.html create mode 100644 doc/src/week42/._week42-bs093.html create mode 100644 doc/src/week42/._week42-bs094.html create mode 100644 doc/src/week42/._week42-bs095.html create mode 100644 doc/src/week42/._week42-bs096.html create mode 100644 doc/src/week42/._week42-bs097.html create mode 100644 doc/src/week42/._week42-bs098.html create mode 100644 doc/src/week42/._week42-bs099.html create mode 100644 doc/src/week42/figures/simplenn3.png create mode 100644 doc/src/week42/ipynb-week42-src.tar.gz create mode 100644 doc/src/week42/reveal.js/.gitignore create mode 100644 doc/src/week42/reveal.js/.travis.yml create mode 100644 doc/src/week42/reveal.js/CONTRIBUTING.md create mode 100644 doc/src/week42/reveal.js/Gruntfile.js create mode 100644 doc/src/week42/reveal.js/LICENSE create mode 100644 doc/src/week42/reveal.js/README.md create mode 100644 doc/src/week42/reveal.js/bower.json create mode 100644 doc/src/week42/reveal.js/css/images/cbc_footer.png create mode 100644 doc/src/week42/reveal.js/css/images/cbc_symbol.png create mode 100644 doc/src/week42/reveal.js/css/images/simula_footer.png create mode 100644 doc/src/week42/reveal.js/css/images/simula_logo.png create mode 100644 doc/src/week42/reveal.js/css/images/simula_symbol.png create mode 100644 doc/src/week42/reveal.js/css/images/uio_footer.png create mode 100644 doc/src/week42/reveal.js/css/images/uio_symbol.png create mode 100644 doc/src/week42/reveal.js/css/print/paper.css create mode 100644 doc/src/week42/reveal.js/css/print/pdf.css create mode 100644 doc/src/week42/reveal.js/css/reveal.css create mode 100644 doc/src/week42/reveal.js/css/reveal.scss create mode 100644 doc/src/week42/reveal.js/css/theme/README.md create mode 100644 doc/src/week42/reveal.js/css/theme/beige.css create mode 100644 doc/src/week42/reveal.js/css/theme/beigesmall.css create mode 100644 doc/src/week42/reveal.js/css/theme/black.css create mode 100644 doc/src/week42/reveal.js/css/theme/blood.css create mode 100644 doc/src/week42/reveal.js/css/theme/cbc.css create mode 100644 doc/src/week42/reveal.js/css/theme/darkgray.css create mode 100644 doc/src/week42/reveal.js/css/theme/default.css create mode 100644 doc/src/week42/reveal.js/css/theme/league.css create mode 100644 doc/src/week42/reveal.js/css/theme/moon.css create mode 100644 doc/src/week42/reveal.js/css/theme/night.css create mode 100644 doc/src/week42/reveal.js/css/theme/serif.css create mode 100644 doc/src/week42/reveal.js/css/theme/simple.css create mode 100644 doc/src/week42/reveal.js/css/theme/simula.css create mode 100644 doc/src/week42/reveal.js/css/theme/sky.css create mode 100644 doc/src/week42/reveal.js/css/theme/solarized.css create mode 100644 doc/src/week42/reveal.js/css/theme/source/beige.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/beigesmall.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/black.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/blood.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/cbc.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/darkgray.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/default.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/league.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/moon.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/night.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/serif.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/simple.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/simula.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/sky.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/solarized.scss create mode 100644 doc/src/week42/reveal.js/css/theme/source/white.scss create mode 100644 doc/src/week42/reveal.js/css/theme/white.css create mode 100644 doc/src/week42/reveal.js/index.html create mode 100644 doc/src/week42/reveal.js/js/reveal.js create mode 100644 doc/src/week42/reveal.js/lib/css/zenburn.css create mode 100644 doc/src/week42/reveal.js/lib/font/league-gothic/LICENSE create mode 100644 doc/src/week42/reveal.js/lib/font/league-gothic/league-gothic.css create mode 100644 doc/src/week42/reveal.js/lib/font/league-gothic/league-gothic.eot create mode 100644 doc/src/week42/reveal.js/lib/font/league-gothic/league-gothic.ttf create mode 100644 doc/src/week42/reveal.js/lib/font/league-gothic/league-gothic.woff create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/LICENSE create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-italic.eot create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-italic.ttf create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-italic.woff create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-regular.eot create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-regular.ttf create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-regular.woff create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-semibold.eot create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-semibold.ttf create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-semibold.woff create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-semibolditalic.eot create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-semibolditalic.ttf create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro-semibolditalic.woff create mode 100644 doc/src/week42/reveal.js/lib/font/source-sans-pro/source-sans-pro.css create mode 100644 doc/src/week42/reveal.js/lib/js/classList.js create mode 100644 doc/src/week42/reveal.js/lib/js/head.min.js create mode 100644 doc/src/week42/reveal.js/lib/js/html5shiv.js create mode 100644 doc/src/week42/reveal.js/package.json create mode 100644 doc/src/week42/reveal.js/plugin/highlight/highlight.js create mode 100644 doc/src/week42/reveal.js/plugin/leap/leap.js create mode 100644 doc/src/week42/reveal.js/plugin/markdown/example.html create mode 100644 doc/src/week42/reveal.js/plugin/markdown/example.md create mode 100644 doc/src/week42/reveal.js/plugin/markdown/markdown.js create mode 100644 doc/src/week42/reveal.js/plugin/markdown/marked.js create mode 100644 doc/src/week42/reveal.js/plugin/math/math.js create mode 100644 doc/src/week42/reveal.js/plugin/multiplex/client.js create mode 100644 doc/src/week42/reveal.js/plugin/multiplex/index.js create mode 100644 doc/src/week42/reveal.js/plugin/multiplex/master.js create mode 100644 doc/src/week42/reveal.js/plugin/notes-server/client.js create mode 100644 doc/src/week42/reveal.js/plugin/notes-server/index.js create mode 100644 doc/src/week42/reveal.js/plugin/notes-server/notes.html create mode 100644 doc/src/week42/reveal.js/plugin/notes/notes.html create mode 100644 doc/src/week42/reveal.js/plugin/notes/notes.js create mode 100644 doc/src/week42/reveal.js/plugin/print-pdf/print-pdf.js create mode 100644 doc/src/week42/reveal.js/plugin/remotes/remotes.js create mode 100644 doc/src/week42/reveal.js/plugin/search/search.js create mode 100644 doc/src/week42/reveal.js/plugin/zoom-js/zoom.js create mode 100644 doc/src/week42/reveal.js/test/examples/assets/image1.png create mode 100644 doc/src/week42/reveal.js/test/examples/assets/image2.png create mode 100644 doc/src/week42/reveal.js/test/examples/barebones.html create mode 100644 doc/src/week42/reveal.js/test/examples/embedded-media.html create mode 100644 doc/src/week42/reveal.js/test/examples/math.html create mode 100644 doc/src/week42/reveal.js/test/examples/slide-backgrounds.html create mode 100644 doc/src/week42/reveal.js/test/examples/slide-transitions.html create mode 100644 doc/src/week42/reveal.js/test/qunit-1.12.0.css create mode 100644 doc/src/week42/reveal.js/test/qunit-1.12.0.js create mode 100644 doc/src/week42/reveal.js/test/test-markdown-element-attributes.html create mode 100644 doc/src/week42/reveal.js/test/test-markdown-element-attributes.js create mode 100644 doc/src/week42/reveal.js/test/test-markdown-slide-attributes.html create mode 100644 doc/src/week42/reveal.js/test/test-markdown-slide-attributes.js create mode 100644 doc/src/week42/reveal.js/test/test-markdown.html create mode 100644 doc/src/week42/reveal.js/test/test-markdown.js create mode 100644 doc/src/week42/reveal.js/test/test-pdf.html create mode 100644 doc/src/week42/reveal.js/test/test-pdf.js create mode 100644 doc/src/week42/reveal.js/test/test.html create mode 100644 doc/src/week42/reveal.js/test/test.js create mode 100644 doc/src/week42/week42-bs.html create mode 100644 doc/src/week42/week42-reveal.html create mode 100644 doc/src/week42/week42-solarized.html create mode 100644 doc/src/week42/week42.html create mode 100644 doc/src/week42/week42.ipynb diff --git a/doc/LectureNotes/figures/simplenn3.png b/doc/LectureNotes/figures/simplenn3.png index 1f1d25c42ebaf6982bfc6d796a84a8db18cf5aec..a377fad3c00d3a8c4f9233c76ed097652539dc21 100644 GIT binary patch literal 3751 zcma)8c{r49`+iE6EW;3q$c&+^&DdhHH^bN&WJxFtlQl-eRJO!q$(Aj9BSo*Qq4Zj6 zlzk~lm=I-+P?n+m9&f*|<9Pr0j^jIy>$&giKJV*1@B6%-=a1)>rMV#wrwAti06a!G z3?7_|!N-k*1svfq=fc5B!VQNv0|1IN07OIs0272FW&t1=2>|o10D#T{fWrZ~&#W{7 zVE=}tsSS2_cNcI1{--z(P(bB>$JKu?|Lg*a02pG%VhAh5A^k%DpvA?7JANFzg4SLr z0)qO2a0~ch%waJ0iUBP_K`VB43>TN7ii!`I?@=_k*j3pfry*btffx%2nDc7#3S)&q z9I6e4>OetIM8t+uo%7&9tgx^(AD;;~H%=5M3Ti?$At3UPG64WQ0f4~CX#f`g2LZwV z4>>t^80_4iCGtG-PmPd}6_|rprv^CwF8l2ZJnIkZce3Y2O!^<2fAl>;!&ZWV zHXAJ5&#J3#T5XI1qCg|#qA{|9FHAy09y^Z2Kx*)!6DkgG{h8-k9BcL z120q3Q_}pUS=U>aM(gkH{?XJl=j&UL@FXF;G`y;6Oij(NdZb#}ML8pb-qbXU!B7(F z5=weYvR`CpWVBXQjZ&$#gj7OMevq&4T?*v^mD&(q7Ov~B>k#YU;&Pipd1QLk6ucE4 zA4#M2Vlbra)@*5OX+rXz*;qWjfl8%kWTd&c^wVfRy1VC^nkX2|Y*SNfM#gAWRV|fz z*Vp#}h0;r-MdIj*?z7Ix#O{yp1&8-1A6^gM{Ze)uP4>FaL9uVNl1DnhiI5cx(n>l5hv=M<>Y-H^0^fX57 zY;UzNc3zb)Gs)B>D07JV`qxGYW^sHX0?s55YzHu#i{<2Qa_#nfnR50L!_{fKacIrk zHgjmaoB)5`-jl@K?kSlH*i>e0xKhahJB@FqR{vUo?zD_B7cqS+0}}%ewli-9OuSm& zTM4JPa)BQq@f zEGOi^J~nTLYcN=rYJ}0V@%s4H@v4*cdodjtOp~oE9+uQfXZJbFVSdCbZ`SdV+M8#v z9K3Ga4HPuj=aF?I@q@|v#^A5>&VCC+zE_sX{t_$tQLfWcywGbZPkc&S-T{}M^oC#Q zz5Y3746zZ@mFjm*CnU&UJ3npt=g;2d?QO>PX`uT8BxtiBMm}N2B(&1OY=$JBATX%v&@V3WqmGaOPjp@mY;--r+Ip7 z$!9i?V(wp9Y**{mJ60$O(<2ZIx0fERUp}+KX57T-7+HdL$L^^+1FZy!1IsaO)I%H| zM{hcn0C53FOm$(xqtMP1d_0Ai&O?(Z2_cN^83i&+XKMdvOSm4X0d6Xkx|Bq{wI8Y^ z5pNq|FO}SNCZ$+u_3sOvUbB<-_6=B>i+56r(gmI-v)9DgY&Cp>KiqQsIwjlC?-i^| zV{_!`e?=UB5c6PA>N&nAB=AoX+vEtHISe2xi4vdrL8O)_V$a=u#4f%OYDi**edi zrY)gzmsts#H`CPL<*q@ijs~q|Jh8{wpR2JW#7ULj+``gq=V~oK28%qoty8UhRx3or zbD)k%lergmyvwJ$(o2WdCuoRI@f7a_i`pmfF$)U`xXe_@-~ zjjF^Ugx}c3Z?Q??_X51FmX6P5!46CAXU=X8`XpatTp`Ft_c`}+n2oWO|8Vw3jrRnbVOG{^gtk_ zdH4*LTVWxqUrDm!W#CrX=N@g@i(aIIhm9j!48BcHF2&D0MU2Wu-~LFf{=Tu%6xy2T zzWzY&QJj?2RyHA%4fW7m08I(4OQ-3P)_?iv%_&~JUu1wzmC^9M6VI-Q3XZAL4{>w~ zjp-f3g+J7vMHPQ$^=R~7D4u%inRgBvJta9jGhY^+sG!(2SnoJim6;ycUUP+(%uEzj27p>W}khGu&Q}h`l{Vy2PEne*V68sP>a(lkJUT>YKHF1tvWjCMUb==HmkA z8}D^?!d=|r7QCBx1XtyM_A}+aBHA*fl`EKs?DX=TS$6QH-TwOqmSp4mg&h7uJt4^`1pHL!s8QJMpAs!HbGXSQ zTFYhE1XqUEkM%s6ilhoV zO+GhMG{aNFuEG1a;^xEfE^?J>{D!N(GkG#m@da!+{&{mjzO$mYTKkstdB>wv^1M(= z9Rrs&3p<1vP&|lFiXON;qqfqEI5MQN)O57n#*#dIXsJStKcQS_x9v-{WCb66aFBkw zz5y{la=3+L`=s&dK7oLbU+xvwzVX}j*q0-yfp{%u5ue{{O@98~V82y@_}eg#T~*y5 ztm%QdrX>P28F?kbZmKa!Lez%8dSFrZ39ahkxnIC*xXSefw#;1^%T-xf7Tgnn2 zTz=kT|8+ij^D((!|6$RO6Q}&G-zL9%^Jc_jeF(O)LfRgz5+xgXJBtQwa`^ZqMb9M% z1jio3rk@nL*VsP&9XHaq^J-0@!caq-k+to7WoP}P*6U@+CQh|9D`kz+GL)M(S#*uCFPzgvRDCpRZT0Oox7=9~`%c)7W5~L%fR9;V2_tiYu!V8QzdLjW2Fn ztd^2dOMTu*DGa5=7o*EK_>m^&`tK^|pPS(V7+UuPJ&y9_!~eTptLO`=~C z`uoSLW5Wy{tr=nR=-Fi`8qMfl^qXz$Y-n6a9z43w9;=!b>k>8Pbs7m1(N?O@XjjgD53Wn}sFkjUAA7gqV} zB5W&;cl*WQez?BycibuU){9}!G`b^i>;i0j$D>?O=)GS$DH7J6+BwBnpjGlS3lWNTe*``lo*peEr>tgwX#^ICr|u86=$ea{`Iz>k&k{ z;2ZGYIoyiVkswE4j}zcdK%eyp3?lmb0hSm81(c$ywVGoqxD6O#%`uPlU8DXFp4(t$ literal 525739 zcmeEuc{r498@EKZN<|cr2w8?wQFbZIOj5|c6_dzNcE;XQlI)DiGKwK2*=B5GX|YXG zV>b*D8v8nf!F>1h)YJ65@ADk*_uu!&=Qx<#J#*jpbzSH7JJ0iXP2}ZEy1O_9IGLE3 zcIoS#H)3MqP-kM=k-)wKIKvw5;l{+YP19Lh`?9{aw#a1<_Z!ZxP$s5h@ecO(`}Aaw zHD9}CZ{OS|BgN_AYxLm3BP08e#`?zk_WFkU5A|t}t*wU->>1u}@`fqvN(1KB&PO8a zC34L#j|YFAHBs98IqDS`%ZEVF#__H!rdLUwBaR}QY!7%6xgLTK95}!e>N>WVz`UoL zwV6+fS(&prW83vy?cQ6O(?{8RsGcf@JDb*4G;$>AAkZiqy}} zI=PMS#H3{B{KxNpx)n%!lKq$e3yo)5F{d}hMhUgM}H#45I3Ol-%F%YND>A-h>3 z#Ikx1{$ai)x?@LF(axP8c=lY^KY8!~8l!MrZ{Xa6JXw4DVYd5S=q1+33z4_kjt4tP zT{@{4{@Oq7zyb5QO&ZN&*SE2jojXG$H)u5YXBsVpe|uZq+xpug+)UXsRtv*^FK+`& z`VeZaf5X6lNg6n2XJXkVz{Cn1Z3BMQw+a4rth4PH)AsM@nVFa(oS9gDeC87HO8&hsj-YbiWRj^Mldnz{I5Lrwkmz zpkCKR{9vwbp2~h|qTfHE3>?!>%ZrM9|A^NuHBobe%OcwD9#9cQx#M!jMb$Y)L_}0Q z9B(Kao!9wsIq;vF=uI!L+sg9tzP`S4z9;3}J)GoEoIQJ1{6 z{>Qas3VXCiWL`?lO)T+$9hRoZ|DVeLn^GCsjvSbVz{(SBcK+`Qu#E+;4=9aSmDQvD zzbnl!jYZKFQM;zSK5a%??-4!LbUJy*A^CC__I>|kVVa8UdsdTn`93)-`kNUZ6rF#g zgf})u83xC14en9R>qv#j+ws=Kn*Nh&hke)!=%weZ_(`qbOiaOY_`N%-6bXhs^esaFncC0j;6|3KmT`9YYt`V0UL>E2{#EJTMPECxO~z8k%PN_Yp}GjZ zf4s35KXBv0IgwG_-)_8TRU<6((Pbrz#@q(PT4eY?xwANbW)^iQQ4Ol|o6pni=1iLW zlD@7umfaVG3h1ht>10@u-;Eq^2S|RoF5<)Q1JE4S2$K&^u^g0yMenVgJWkp=d0n9Q z{NFzG*T4Vdr@{_}84D-c(oW>^tySKPD|?G3D!TkOR>rB3K$^qG0Ye%(v5aTgMfOO} zFUaHfa{CG5QsTS0SI}3q|9bN8v)QBPP8TqG$Knm5~r_2lS>yh@Neb2;B!?yjUI zi?K}C+YtPIUvQW7I>GsO1!GvO|E@m3{gS7$Ax&V1y6B6G{bJd0L)!oPhmhP@=DhG1#BxeGXTm=-_EBfz?-KcRam3yz{|4RS@{> z9+-p7>9C9XhsnoJbE0S*9m>J+~-czFLyqw=384f ziYEHFp^6Pk3>JVNi^>V9(A4f-%W|=)h^lnK_2mBMR}`4?}{8 z3Ei#_F1@%Yl{a_Cs19v(5CM&P^VOkX9(_XAg=o zf`PD8e7zMG0*mop1USCRfEejDU>C3|o42kAtjrq!s#9T^wnmZn%0v7JA$t+J6k;3m zj|O(kAJ7-@mhf?A#xXSS3u{uxOl!-pN?N^TP8Ti-mE>Pdy7b<3vO~8t`?H-TBEKlx z@@lq}TJ|8CyGK^}5u!0@LZ~As7p&Ir_iw^Tp$o&i?f{5U7}?M2^H=cuM@1X77T&rO zOae9zIpSr|U?QBb4{BTju|WV@-QyvaekiQc`0`TWoROC9iEsCMl2 zY(}3Dcc0naFr~`kQ%B9%RzvJ-S;G_|>Ey*ZPFR#xzI(k2Ug)rd@bxtX{Q9#<3O_{{ zlRd1cwrOT_su|Hh=pMSn#W-e0ET|bk4^6_9$SaJYN%JX(td#>K4=bwX0I6q9uH4MF zaoI7$;ZE@7%9eA;p5q@^Z+s0kmhO71F%G{J|4IY@fBy^tj+%ums7JsR%QC#mH}F`t zmL#|b9S_dE@>DWq%z2&-KSOSUO_=T`T0HfDxnUpr(o!|_C1V$DOZ{!QMyeVJI=g-k z&A&XLNDzQ4!ASUfJw{xa$;6zB(K8G@QaPMeN=a!&$kn&(J5=R@Wj&3RR@?)SElIaf6U9pjOMh5(=V>m*0 z3m=p0&2%#+(M=hBGrX|I zOs1w^f{VBME35NzioeFhY81+t{E>R+LeM)_^u?hLW2_m;88SEsgAlJUcxXmD1b~bo zJ)BiLJvzX+T8kp{Ywkt`*S59~75|-H6=SjLNa5#J2dpgJM1PLZP|h1TU|1*$NQvP>cx7+<(tkQ z<{w8~F?PM=z!N)Pk%OP_KuMq-?IJ$oE9N$WxF#nx9KMvj^u!Gg$`+Q1b2#9tQy^}( z|Cm4hclZ*<%X;4+>6H+&Vrkn7Dz8V9h$Liy_*Obs{7EgLsA1vq&|C<2%goL%{dN7} zVb=8>OG7NNwt>&^q}&b{-Aqi^m#HqGk&ntlpFdbtmE>L)0eTZhH_- z@1MhJGacFKh9iG7`xDGTLkVs)ku@*VLGi0nnu<9Ewg};kcOlU|Rs9|=@HN8QfyJeL zBoixUx17zL(lC%9zJOmpxmB04<6w#ms0{<1buo)3a}m1d^#P43*t!C%GLv=~&1G}~ zjOxjr9ZUCfo<#?w7=6Z9(bsoMP1N7ZUSe&TFP#v=FTdYv2qZxIF&TAgA3%p`*R>2| zAvn-o<3MLqUBHM)^*a{e*+yy`Fu7qIYw*2FSzGa$Pwg6SBF(_Mrxr0C|L6ht(6!_Z z-2K48-v*#%-@%tR;GHKl?``NzOE;Pl9J=*Jmh?CRTc~1wKdYrjuNpQ7k_dDzf_2(c&C8c^&5s%YK&Ih<^QB z^kx*o9v}L6f+G$b{%t(c!wPENftCSlRO{-=>CJ3@xn`{0zmT z3JW7$_I#3I-I&|EIIza2i4Jj%3W1L)J2!3NpY9?CrB7F!@Cj5-a^^yfYj22cI%Jj^ zdxxli7?7x%_UoOzBz!S37z&2eSsXX1V%fzy2vN%V z$fjqAoJju{tT7MZJ6D>ba#P0Ig8xUO$ULg{7Fxw&Z{8S5iw~$;Axd77?3Dw^xtOx0l*RW)@F|@GWgm(?7S}=Ec_8KcMt6E?ajLfis(6e+pINOSN)hl z8qbVvH^P_T=U$d^$;vIVUWadZX-Xl_!=E&vZiJnzY_}g)!Gq@fsW3Br%QJfUm%9nV zLPzE>66`IaZp$uPzzGRC5?*K|2`tGuoFZKRiniIMn~(e1vhirs^z zPrI!=wkfkpEkLqXl)(b{S`jDNJ_jdIvQEemp}sTa&{=ClE`A&Ct+$Fv!0O9{bV%^4 zTEI88ZP~sY{HC^^Sb*xT4)_-!_lGeM4{sCsb?!0xy=~0xA!6PE%J#YmIll5KF}I#( zYjak2yC+DKmwi;B^3H)~?p6{#`}#rxqP?2sNN1O@vq{109V%oMLxh`Z|G$BeG`c93 zYL81ZF3bZSzmOcS#beHQp_BUM)_AahzfV4vhhkORsG!s5al05bx`p?gXhJ)Z$#=xE zDIw(cJGkR5rVk)E{5&6tGPd>1z5MJyhy6G5fTLKQ+bhPPiXv02wNK>B{WUGi)!lig zX5*!k+6X7KuqiJQlglf!Ejb@F0f0mx9xIHr;(HbB?M&1xf+%;9g+znuK51NMMs7%y z{Zrt$Mt28~eD9rLbO-e`{?)e%i-!V{Ww*y@?qKJvd8B8y*x}{&*<2!CSY?&aff{4AQZ9BFz(-z{(UX}2ryM+fiE;fH6icL<6{sT4 zQro8)&>nkvC~p3%!mhFMM()};t1#IA zv?~(UQZpkTl499!baw%I<8Yq-%xd>?zI%D*D}2v-(*{i5c_?7Z$CA+`e*tOW=#4l% z;>@yT{Z}f7VK`hrMF8*^KQUL&V&Yn&CQw;IO5gTe1{DGD5!7g6;Lh>SBvWFLj$T-XgrtVyb zCHY$IcooKK19jox+?fw;fI{4Gj}Ef6xewgKKcPA{ZaijXjMofm8`nZF5iJfv-58#g z{Uo$dH78&bbwS4Xuph|?kK@%;i|SiEYz2kdr$k|t^s;^>Cu zbPAziP*y_}^HM zWEhQv9Rq;5FSQzvcK?bKVK4VSE;jEUjj1v8oqAn!IB`f!5mtO+H{u3uOyeyE`_Ly( zjkZ53cWly3_QllbxANVC7oNJ}IB$T@ZgP;i4%d{nWIr+`KKaLrlLB2=KU5(<>|l6k z8Sm|F7Yb)_aCELWMQ?vA-&~T7k4TF6u@|bowViipnQ9~gc_xr#U#%pls5?HN4cgu# zeisp@vVMe=)kmay!d3Fxi~lX%`a&-_1XBLk%iw_GyqWJ36L-`ku9sQO$_Dw}GeMo3 zY~sDqZltykIa8v2H08t#kF|F}cvyS&$?Q~OQva!h0emH#u?0{sSq*#FR`o8|zUPS`Kg1j>ySkIqr8#&b*{9sTjBZWq%_D-YhxVI0d%a@}Vq zg2qIx1l-y3+)n4YJk+whr38*LyRh}a?#Sl4?^naZe0;Pngph>WtXkHy5ji#wpd);B zN^bD(0LCo+*W-Yrq(^|H#YaD0Wpr5eEZ+mz#q{fvTp~|h`w%{_;@3}H=<-}g;xl~5 zY17f@l-A^U#r}OBsnhL~yYQ{vpT)<*KdKI&aUSN-g5Sgy)$5~OzgVFIuenx^9?hei zC=-4ut(T z%q>X|rL!Dj3}rMC8J2!EX{?$UCC{`>6N!;ajB!bW4<$^WxD(^`(HK8+_1?*c9w-2d zfQ86W8+{)YY_D*d0XdA7+q^RE>`=9}3!(GhiYEWnA>OO6n3H|%;YRWo!*X-M z{ErrpcRq_avsEAdXDEA{3&3i#)VJRl0z>neFzBu9r$I)k-S^M~y*%h?bLEsaKBjW7 z@?}aLzN8!<%Y|wzm+gpymgHL#Wn~kM2VG>ZnGH}@hjtky3x*d}d7Ix>DP(Yhznj$; zdRM`ka`!XC>9_^vYD>j*yOc5YwyAzSZNoO~VYNNY1);ci-grIICUR4Ex3m0;a_hgX z=706SFYA#qFn!EcOYQ_B2tIwN1qy#Ia=DjTQ1)Bi%5CS#n&Fd0b-L2-eHF+4dBk)a zut9mSm%%dQPB0JJB?xDv`6`5-9Ro!g&aW>@0 z)3>N=xomS;e`|>Sy0u6R-7qP{;|dIhsi0%OXia)ey_>6@lWWg!$)#)Gpn)&uhMeSM zL(Zjbq9)q8iMlfX5UWEqD*$ZAN!h&S#UO%gZ2j|Oqrt7n+yy^f(sM*4rIamfgL4CY zQF;A{^XyQA(@jeP^Ye9Xwf_h1k!taKYKi$`C zV7UQGjw#N})i7OcH!gnpm3daVe10yA^;dxV{pi2eAN6Fm@h^LQy^D9{(&Ug{$=6?U z=jhiKWM<`ACAI5)>OdB3OgLY*5%7)oy1bpH{?CXEPYA?;o5L;6k+SO{=pC`ZsGxC27Jad&r(GAgxK!B5_Yj1go z|5%aci$LW=VX0O@hGB`o7wtcT^P*YRDp8I5k|kZA6mwZg^g~pb*LYSZxk(j!@g=8; zqcX{fvo7-98Xtu7;l6)I3de4QIn^b|3CUa?k|<&Er*z}xPDlCgvHl5C6m)?3J*rH; z$nw{!iae7xwEUN0?><>?k?I6 zPpUi9z9m=HuR7zMV|A%!1;+;OBzW~bGjhhOUbyyKKPK$uegI1gsOtyJ=)%utBbGmo z8EsC!Z~si)g=g4BThC~K1 zX$7-k>=S(9oIY}EGTi~`fFgi!MtAt(8U}|gOlRYbvs3C46|3(*#=pM53<K4!0z7|7U^@XpVVG(Sn?F1<$6D8|1rR4uh&C6%)t$5xM~;T=%9byJ5F@a^7Uj<) z-tzhn@A1*-w1Fvfw$0@{GlQzh+j$f5U;Ly#1gY0;Tp&ft*JRBEyoTOW)-8!$_$y$G zJOk8{!@2ht>Wq_O_S{Mk7#dG+``#!Zqp#pnL%9wBT_;i@DerJky@J@Ly2#2FD~VJ> ztC^2puGwtC9pLh4OW>T8Zhm~0!JW`WDRPo!Ys*YvK4$rwQ#^n0h*MV?8X8(MHly6X zI6QU^QQTYTV^*cR@c(U=eM(8T#)BkG<|iNE4nReXlDW0u6eXV4^Kum}zct`2%|fFxM?e z`*As4esHR1bzS>I6_m^V&6#_H9@B3D-^59M+0w2wbtBcY@M}sgvl@=?ll{Qg1|C)r<3sFjmzK>7BIuO+@s2!NrWLw*_dSB3n)X{j|C zIRiQ-j<)VJ3aD{#z z{<++)ndrXtE|lpmG53$AvSyuTUpM})l;>Ia3X=k`d}|vsh{!0=m~25Jm(OHK&DU@w zzAzAmoU$Uyz+zI(@2DEX9PuHEc@Qi3wFf9EXm)4NrL@6BCh?t1f{hJIo(g!cP+K!zxRGj(I#&?)}p!n;Ao! zf;nS(d8g;{f9ceZ1->UXjJSL2PV90}c>DM)H2X^~q-~P-;|h3EHXWIqvDpe=;eHht z(>KVFQi#k))@}ba>x1%pcvrp3)Q(X<2|vDK#N$U&A-((@Nxg)%Rn83d3Mp;Ce-Qs1 z@X_=Ri@K1b`@Ku)DW31YwF7PB^jmtf#Oit!81XF6$lnTG6mHVB<$4P;PY-Q2`6kO} z(0@mG>~}+{FQ?R(b|mj0`%{RuG|tnRt@qoB|y z2LLuSZI}Jl&oOKAv9o}@ClHd$Ysf7`2BV=YzC~k$Lo5DP;CU|h zRQS~LrW+Cc{KYQh(G|b0ka3PdRF(ksKN%h(ADCr&^p)zwqaRP(ZM8*H8?#WsP5N#e z>ldXlSofs>f=U*Db#Igf^;daS(+BXNNrMz*-(Ojw#gW%uI@Zf?U1cqXcG=UAy7tOfgJtDpk4Yj9u0 zU#{*a!o*(~PM$***v0d0@TiUyfwnJgojCaTE*#}lVAbu;4fi>D%h_uKK6ewhF{s&{ z`-0u-$8u>7^MtVdk^|6kY`=<++1%)mje+)Tm>D1u{y5a;ESjnK^wx=3db#k6d%LsL z{y2lFuWFzJlL=|U7bu?n=knI}(AGQ!OKI%?0nUyc3|o`Qma$8)je_SjM>k?Wj47zD zoDp5x9-6Uwu4>hPPR;MDEU85MSE4Bbr@b(Ws-4RflS%p0*!->$mI06?_-=Wi1 zUo3xo)Qe~)r>1~s$n%&cGz+WxO!3`il|;>#xFYTp<7~pi}8i#9yk$ zv2pE5l)%6U@;Gq3Xp?0&xZ8-*88an}#zZ&T(vYO zu0R)oJ*!?lE%;r!{2Lo-+x|i-e~J@u-H#n0Ro+2bPnV-;(<(i2I$m*?ZjKM7kU(wI zC$}zAUmms^@bGP$Vrb6wNz)C1=s*5fx6U!Xz6i14+*9$$uo~#jkWw$O`=_~J_))sk zyE@YFy*0Hpw)q+7J|(J*I=R<)fR}PqpUXwf{E2bv+3h*r$O_1NFb`14ELL%t|dNSiLHX@D#mns5~|#4P(f=Z{PTl}(A@#7A~WlcgvDW27@D-Vi5HlTFc%jHqg zJgZwkv28&up@*$jX7GB*2=1W@EKe1F#v*Afb7KlI6Q zi1PR~+Ws2@M^6`yy33Og<>*HZXeDqe`n*nwotpFF>6&s&JmuchU$g(N-xtJ()oDOO zWPRC;wtv8ADXTqs_Yqtoo6}nfXExjbHV{^}_U;DWD&@k|zb@o_*I|5^HrCk98uHY; z6p-#0&I-1KD6_TpvxEV`Cm~mob8~eN91k>Tt<~B2`$w!GwYA;_WmAQCcieLaQqUu> zCVha`_S5EmO#8`^VFg*=141=rHa&M?eg1HZ7YaBGD{(q1BQawGUwx|uW>hHst0)?0 zLpR=@%^T>SH_|la47}lfSy*{MWhJK^bub9<#++uu=?C$Vu{)5SxF91b$e*3%kHemIZk?Eml0VxtbRE~f zq9K>t*e(*L6q0aBcxzZhc8a&+>?RPafwpAb$&hE?KYd`#dcl@+W{T+RxB zS7ofXK^(L3YT@k`x%tbv(6iDHjieh+zdf{iYWx*N$bg?XRHAU7E!K?E&=UL9B$(#?Vnl_TdnV3JcBF-is&i2iLoR-!vqTPhZ%dN!BKJ3P8JN=5C zKXp4BS)VLVn5Z_~7*)?NJ~eyJS%qqCZUkMTclq$H)Qjs8n_ftRxh=hXb5oa&<`Sv4 zq=6`dO7Y&@}y6 z`(xtNSNk+-qb$JynWuHXr^si%sOui`4=SOf!cV6Uo4J8T<S|UZ)#3g8~WD6==ft&y+80568}8OpaTinsR2r?nLWxNls>V-e`J#G=1adTqSp( zcmk}M04_YnV(pWK?~%}05-*%?82f^^RgsuWah+?2War%7RC6(llr5MjQBO8`qHL)A zjaXw7lrocDxw+)C6Z5RuhiNiCs-3V9jRT4kjKx>QKV}6UA8^1-u&Hpli-Uy5#o=hjmSXsv|*_%N@x0 z6P0r#LzM%F5f_1s6AJrS6!eFmLZ_v4dm-jLkYVwcK4K_LF8sdHaO>lbM!9^f>ISu!K8>9v<;i0sw(y4i z;Z3RTn^pccQwFN+!_{>!i zQJbsbMqXquSC8n)Ceodv3)wleB3xv)5_ATkL7mWyiSD%p#pyq=H>b*a` z`*%Ik2@@{Td#gc?oz5`Mo{C+@@O$Hi5Hoo(Q0@Lh{P)HhOD1;XcUUpceX_~!zY1im zs72_^8Y$$ihy3D)vXppoEl#7=bQh{UV3iP+1uX-rzc(8Ui{GnJgo50F#PY|(;z0D+ zf=ibR_ph?^!`V!zov1SpPy^V@nhRXAWz&bXH4IA=L8Yy2t1Ja8-^W4-0CP zjFq?IJ6|boc#$UGqI5M%Q`kmI;@@Y4)#sY;`n@e5cAM*Bd5S!DRgXqv>dd_~!u0#J zgk!g+S<~~C^sKhH0t2RuNbl4Z6%&@GSyxYxnF9l*^4ee4ZHS+)0tKHUKMH%<63qJp z(l;~H-OR8A|9f61x%j8McB`MPGv87D2{76VGI8JDx;-zrPa4W!Wn*LOOv$lD-uzl7 z7OEM{5>7;`IvbLrA&Z`hF;W+|QRXPyk-dc1iJVI>@x5{glKapbHK*sLo5klmLYi#X zj$Db=tG0SIH%sO_)Fo>)bkn`7yF!26S9@gi&GJyeyDZ1w|+E+}^X z@Mb`*mpxEgedIFZ{T7iql$ve)8XTmmP#XvEEUeit(&I^`TH;eCF}< z$<0#MM8#+>2^;@hwa^#lCT3r2*d39YC|kTGl_FX@4~Nv~wRG9I1YtIEtJ0#+gxm6yX&nZ2711DnCGU!U)d~xc2%$<)S=fOE6>=h?|LsC1{<@@9 ze!0^p;r?HV+V`V3M~4%n(ofNusUaqG&Pwm%9dVn1C|&OYeL@l>=qL^Z+4Kntr&4ao z>{34AXLqM;huYK{EEL*PbmfrQguo94-Xj4B_KlYk-d>j55K&c^|=O%jJ_V?n0sxvw^)m)!mN< zjlBr7a~&7jA$A)X8v@BVh+E`UC35ECf;@Ssgge_8_M~RjBhZ+XjvKYg`yH#Y|6!0h z+JWa9TGd^tW*s@D*P)rFfgtp{FSNImcjifNUYB32Xs;2!>V|2@2MsOSZoZepTf9%& z?1MvpR=6%F09Krw9V}o&xUhmeZTWP5DCLrtHqsB}bv`Rot&E;5+>?1deArnMe)DqN zVls5Q^OZUgr(**S$W8F%hU;<~wL7u4eCKH(J3n?POs-wuFo>kq)7AvWl+#g}h|;zc z6e)wIzg*k#y*2Ya+ zv_!6k=_Atf+P1gA%?kH$yqF?dO7oPyv{}p&<-RqjP3ZL~03G~$a$X!S10*nj|N_JH1jR2woueJRaXA9I-AL0Me=6&H6+Tw)-YFw)E zcf5T(SK@TJX4^r&*9Xz~*(+3gbus1q3|+7Yo?#J-#}o05OmiNze zLqvsMN{*qXO;`lii7ax>a&z$Kn<8QpPt@qN2~xB~g~DRrm%HwmGR zcSCdWhA6ku=QB3D6EdcG3nQRYpWCx>c8_9mYJ=`!#zUJt9 z&>P)(`aQJydaQ#=HH&E>hIYTVuFawLlIq;d%r!EQ76?%c2PFH1`3I~*GV7X7!9%Ni zV*Zmj_2$T)<$bueCLM)s5e|P;#$L~Y`qMYOOjYxpawJ5y8X!)?Z#N_2xfbobQ&`Vm z)jP%2|HCzl@X!lF@2V|B&sN&qp4<2oEEO_TJz%sUWwRq;3dF-P^u7sqSr>5Byyh@F zMI`0Xa(`?~`y?^rsRn%$SH;kNfCBs`9O=jhs?|Ax%d-#mp!9Ah{ZHZDr!dJjuosgfK~@EAQzG-PxH@ zfZg8bJ%Ve5qe}eE<>a^gCK(Y3pwmFEH1C*4g%FpN1}?3mkS59owW?1XpDfCk$x`e3 zOYTIr$|>fW$sx#vDKB(e%Gp@ev0l%-8r8*>BwqbY^54@Liwmg{)^kz4|(YqYH=k1?D)25Cqucfp_%t))<(7IUN0}e*1yMukrFXzfYr=PtY*y6J{ z5i{Z@&=duy(>+U-@yn@i6v2&mHP<<&Po96+?zon(R_3WC!^;xRLM%b;H5AxiCfVP; zV(;8}_H#fmX1xyP-rXO?wg4?)o&Gz!@{QCUUN6tCYw?fjzZ|;bWm2G|7wwkJSM(6g z3iB@c(8x{jJ=Z0@V+OQ_&V<)Txrv9^9oY6KT6<6DZQqHCm=BP{b%gpp*V-0&-;W%O z2bgg%?R74<=XW|{n3zeDfSo1a@MUhsBgNb*Krd*%lbB&ytXOjKvlnjQ^C&w!Di1oc zgjmkt5l$wie-UD}3xoq8EbeevBSznIYf*TO?m9!qPGGSwpjq`#48igt$iQ(sTS=i_ z5r3o!Mfm!*OUA)Vx+F%m^6iS%N(NW z1Pb_vDf*n+K68@FMAiOLc?+A~+nisY`z2Afqp2W1m2h2>^epib_~&;4fN{L|Yj$}? zYtxlq&Ym1ByHM4E2&!691IWFBV3UsnP58(C@)g%Yw2ypVbp{N0d9V5P!!=XO5OpHN zQlC3rDp@A6#!h+8o2H&pDl1wH{S*mo==;vK(zIpGry#a0Ce$Xa;3qwY=v$8ZK=WJe z_&Z|ZyIf2YGU_ttOaGX;Nt>Z|YzZE)>mC%ldQ9-Yt$`NgrP}8cX-~#60*|>mCs&@9 zxf*z2C)!g!_G$PdkIw{s5zHIyAnrxsJnj;&82@MphlR5{8Vq-caLwPrRk9|_vFXq9 zd6)S#f8wuCj2%Qeh$nV(qs8WBi$giq0a!W$n&u8V619*~dX(w^&< zH*#ipX4>qk;oF{_3lABx7(W$~bLcnbHal;G=H=mz9z;uU2%MyTaV({dz7O?q+s7c2 z5N-szG$k0H#{wew61T-WwUAD6H-68*0k6f%)BLbLS7SPAc?ag}?v?NC%1R2PN7zdC zWyoVFc9M!H^Klrl0I*e1aAD`|>UTlCbp90gN^>G5!FF!)S5@^1C~kk4A$Jf^tHY-CM{jKH z$nziUT+cBsvZJAS3)=M)gY{KiFlmEtoxo20iy`#>OXe9CfNrl847k}9ZPemHpc46M zdcvZ9_9Nf)ek+|cahu?q-_cE6e*NNUa^v;GA#+e)z0>FgOh-ZxbCZTBR;fc{ZqNmc zT#0ibL~Xlg!?)!S=DR(-T_7i7tU)aMcQ#S8ZP;OcefZE=dPoqwxD{J*;J$+YE3ZX8 z=l50bvhK87xvN@Ec9`ZrZp6I|*{bfGRG(9gb&)mrHv9K^s=)?Fp3xvPI;Mpp(KDjy zpJdCoNS~%o;0}2Vu72$^`PkbOo-qya#-#DmX*|bVwqafc22o3V!-7buFH~>eXDt*) z+;ktuC<)58S?^1C0_ZbJ^3mTD`A*jHqSqmwJJzXA4)!R!FhnL>_gU#uH-tb>Yu^6Z z_9{A!cyl6xg3jD z*6sIMo{|0!CgPNaHH9tC^<0s>&L{~(H8Tks6iG)k)xaV6bk5F0phL?fO zPd){%&dz_J$kPG0U2m9W!ki=mL=Om+gq+6?Sp@Ac|GY~taKHU%=sUzF3lFMicCrV4PHC20=x*GH}OdF zC;KKdWjY9^mPm4;|kEmBTx$(=N-VTitM8C6RbGYXb{GyH<=AGnx6j zG?q=M3AxEw17Q^ldW9rRl=rrlp%5~`c3pkp0>VNEA%KXC5#21klusncU)jffc_S;6 z71#^;oy#9gd%Q+V96Cy{l`PQW8iiCll*1kL8do;ZjHn5w&xY>Gn@{aTaeLl2m*XT|lfRKtlqk9Xh2;jP1ajnUb zuw6rBHVp{BntYpGyq{~e1z$4vE0I+2^&aIrs7xhgZ#f4m=jMEH*6J1>_O11I>?}jwaz7GYx-I!s8S(F(mldq9-Lowll zl{6vItI~F#3T~_ZO~5Zh;RidQ+8%vfhA=ss9vy%#hSnuwR30G)@&$|EW6#s-4;F(i zRBwKH+3fUT;8h9mEf05tnv(P02Nm@MRD}4bE-j>f`jg$EDZvG3c4Bpyn?*HzOLC_2 z?$_spTc%`U_i1vDZ-gWYlLIF>+`2UvFO7(XaWtXfkj=&(6q`moTH+v3$_5V+QdFLl z_;@Ddm2hoiId^n;%SL)oY$3EtSXK?MC2g#H2kAQ4@j=19&3Yt#vuG1_pYZ)F2Q)pl zQB>@UdNzXm*vFMxau8IB{_S%MTL)?Cne)cL;zu~Ck?Kq@;Uvuncl&LAm3)7;agXf` z2;gsz=yA-Gg)WVt1vz*r{qOU?1{# zvhO{y*KUbk9fJ7R`jPEsaqYp8fMrgL0!*UMwSmEP*smdNGLaKqZrr3|PX!g6@$~e> z7HG!jmMg@uT7=RQqWF_6ATkXUU)~QM!5p&1+`%`ukN9ku!MZ+`oU@Vcdcj*$S7B)r z(5)gHH{4Chjz$Zav|5srh)sH%v0OBTAg$3{U|%)f&VHF6U6TZAIB4>;t9UHfP4t^o zvqWxN(2=_IgOwpmo)6*1IJNW70ArYB7L$OO)qI@^`Cd-~^7s-+=72Xb#Ln}EojJFz zjj3sh^YjramEWY*ymPKJ9py>p?uQ8eP2Gc8ZpfnBEe!QF%q21%Q6k9AfPN?JGlY%u zi&I3~s*TH}jk`xv{PT|EPTKYPBX_Ag0N+Dk1CWy@A3P5st`9P|V5h4BVMm)=)h0lB z3KQ~q3ov3qHEL(dyK;Z{<)0|?BE6F^9Ai11=+e*MHk*RZEHShpVcm?<^R?hnwXlyy zxJAb&dFnLKLb*?oxEd6)E>z?BBb|6?H2f9`OJCa?S@are{yBu*FmPqL&-2q|=X@vy zCp8<^a-BN-u>^HibmP2;Z&#3HF2O{*CRWk(vRp6X zHhd)FUf#LBubZ9(LEd2=_=kxeFeqx3nh9{L~5uOQCYmr_OBS}Q@K@+;?E-qY$% zf<6`Y2@$8d(e;v)jijISuRbv#b4-8_Hsc+(t~)E-1@ls@!rd^V}L^WK|& z&MWtaZzYwCyO9+e;*7If$L><;1#5PIT{wqB-N&%ly$|{n@)j|4^tNQLWNp`GYRKGq z_o*H=)TZzE?_>#k$Gm;|*+3?xhARJBNnV$J7AF^WcecD9?@v6d;gthryEOyDL1yfz zLRbtacYav1C04+tuE5UM@dxN~!604+1atTyS)TlYZP zBtZ|@E$O9P%#sFgtB|z0g8N}` zL(g~<#LI-LTKfiO4RQ6LK^Uq@1>B?r8HiA$N9;T7N7qGyzsZs3ZusV1-!^nxI_B&j zV*;bOeRt!VuvwJ^yeUPs^*eoR9~TSj9v{pwf(_JsF9FM}MJrUkL21VATD#)!R{}6k z@vRjR^~DqUrk}$LUtvydAkve63i3y+ynH*%Zf0!7W=5QB5@m&etsCT#FA?WHa*H9i zCsH2H7GBm_U%TP#M0Jyrhj#F_fO>@Oc9)6dbC6$_yWCq3x4WeVhMF`|$OA&Y+$cfHP;64xkbw;x3atO$5p19?5+)I? zxBGGca(U!KDrzz8q^m2N#H5xA;2AFmGKUaSH4Jt@W}N~nX_o-%G)L5|;_YlIReJ)A9$A0NZ$Q zD{Og@Qi>x)u!gYNSsuAe@N z>9m60Dzdkr;F4l&-QVUk8+nKHl@jhLHm)h==xRhgl3$%}DAb*nh@O+K6yE?>)!4Vg zmHR!By&ihnT<50t65jlz)H_xI>g8YYR&L1%KrG&5+_lSVnWFGv{I^T~hEY$y1%Ai&tox7Ml1fpsnjBm8z-ORMCd3yCNmu3;)0R6^;)~sc zOcfk_An8@pcZ;F;tN1VEIC>MYlB?|Sn4O$ouQUF%?$=;hHGr%Biox8>pjJlrqdk8=u}c!Kr7qlz;ZYe)Pay_uiI&WM{9XnOF12zFQ-v zZGOcNSID|+-By-YtHYNAMV^7{*0>wia)(AMX%6t(rH8((5-kwKZQwUuG;5u4>2ip#F) znZhBq@849U`MUlJH<*?>I!Ng~i+|4sC%k~B)SsuS+{xY?)HCg$+t1+>;C`dTwt=cX z={Dhh=s&($F06*@gOyxmN@}g$V~m454EiMc<43}YYg0mH*6=&A3tCn|??9u%IUfIq zt*?NJa{JzfVQ2{nQIL>sP(T_+rIGGN=|;L?L_`6lJEc3NYb=l)x?_~??)u*0UcLJJ ze`~$VHH#&S_nf`=v!DI!ea@VOMiy&Sn5x}2cpZuWx8-!4RgfVu#U3g_-O{?rQjf@UVc&6P zEm%19D-RX~j`iZksbjES$yh$3nF3+r$F`h)5pJs>(=K21L&{zZ2txtKL>%14I(DVm z!{>qnpWPOG{)N@+N-<&@1o!1gJ$pKX*INO(kUNu>#s&SjX(^iO1P#KDL$wXYj!B%Xmw{dBw#(J$eX z7kX10%HDNe#UFhaUZ#ih{8?iHnOy=Md8##Pu0Jcg>I&COuKnqL#on9g%XzYuZOrF- zIPQ%pW*ZKwde-C0Lbh5f0(w78gY9sS$b&ps>3*7CS$_4Im_o$<(Il(1_lj94|Om1*GH?wws&%}w52*iALBt$pk9X#{hwoJ zDAV=r?ucxlQlUdLJ|26j2s%^arU?yFfwg+NRQCVARFF_c()ep0aLbnJ%F2fG>db&V zS+0LB(^hdN-x1LTz20ZvRS~Pj=PTc~GS8;sydTXrKfPl#6)HTb5w$vIG3GVQ5vnd} z^_tRu=hH3a>g1vPIQ5pN=PEyn;fAamevSUNY8;JwsCfUKzlSUWvy$pCZ{Z_{P!H;R zJr!oFM6nV^=dJvMFSF93Z+-Ku%P;+ij!n#v}L=>na)_KL$z}?xO)bhx~GLt8h zgCBCbZm$#+%tYEZ)S6PqOzlMut+{R`oa&f=dKC980zk43^qS`WbdA9r%h^2p-MO)3 zQD7EgBDwe|Pi(exfzF{>%tv$n*^5=_fnR59{^eLe;}NG&Q*>8BW5%e0cTfE<2Dzq z&OIAU8>&1zFV;I0p+M$xxTH5YE1dVUDAVE~8r$#jDQIcNUh@-ou(7(->E#r@^5eGo zlO`|&N`ua2k{0dkhxvmLAU!EGv6e~qi<_Z*u!))*u8=n-f#}41Y1$A9ep9{U~A+ zsJlo9DD|rztpbc`BK)n=aVuMn9fzIel127PoDi?d3nNo;UrwVohF#>5s?Q94W0eEw zX3g@W{5J1>wND*2hw9oM4m3ekV{!UxqQ-PL8dsSyH5{XNwOyLcNU5|{3fC7m#ZyJq&Ml{= zfjE;jHd+Qks%Lh@$7NTYBK}};El7)bg!518VTrcOe_=#VAG-Ltp<2}NG4oN5{v9jC zP9(dMnH=Up2J!|jtaJ!5E>(irT0whU5kdUxbgIcHs>w$>v86V2c%poF=fuZUUHo?% z-Pwsb@(^ZzC{97~0@%hOB zi}%un_IqFkb7wGnkm~6Dbc^5M%SS8qI8b-SUw0YNlmS*)&|nX`%oD%YeEI~3(xid6 z#BPOOSRQjI+{zT`;#l-JgEBj89f9v}Rt= ze+bxMs^QdZ1*+6USGOUtviQ#>n!wxFC|irs>%zz7MvGkp=a&9Efpje-qJR?V5r((9 z+zn9PsE&L#W8eiH0Ku%(=j_##KT^9Y6ta&RJtyTV54*JnPT*eB z2#|%v{tj|#6`BuQtGThm5-zzWs0VC_46;S=N9lw_S5iQ>TN0&vck$3+&XJ3)#W!)N zPIKmcmb({j@g%dI3D?}E_+c;DFn9OeocLLbMLdVR%(&{~xO`?-T~e!PT8SWunutmX zR#laJdRbYPE^3|1u9mKrtu=%5^d-f?RGy{osrsW{C-cQ4A2E~Ov!*Sj%0_2ZQCIOz)X2L77^>i}rgnpKz-KUW9u(e9@i7(r*He z-lE@bRi|=$$ds1#$t)^pTRizNOeSOIM)oicId%l35X-0fG3W(jGS|)bsOX;Fpghhy zOS=Yoh##y{M$GaQ1ffwQgnL*&8PL-T)%s-uz}?{Dm*mwN0+*=|;P0t^y~sbWbNw%t z18Aqxqm1H$u}p48#5x{iE`q5aFY38e;d)G*K%e|Xuh9;FSck&{J>UJwF&s#szkXRp z3>8C+AzdfR!X*VGOee*(>WX~d5_}^zpo#L%bvP(!hDOnt?Uq{qRDSybg>iX!KGiC#IH<@N_pKO<3w6(X}N z;dfxb+{lMU2v|vn(vk0_8)Cc+D#|2kq`arFC9qj<=+3ln2Xe8295AuTjhdddrmwWMYN||+ z%z_PG?)GTB7HW?RPUBVUjM(+(GM#PW_#*D!swJ6OmzmsYG}d@o5|O~5$~?UOf)ZAy z*Mylf?M%mAL>a+ZvU6JalW*4_QrZm(Znz=CGw|#!_--q8*`+ake>ppMZD7B}(!hWy zjFRphU9tXc?b+$&DtW8EEl8P8Tfc}V>G7gfRyYG-ZA_@UWUQiiQT;@gMu5X)gO|tC z|BvzL3E4ru=4fbsd(E}yk zAKj0*$I>K;x}EU`m)|dnqmtmEorAr^GybN$8mc8xZ|&Y2p^%4=EX~@l(!S-mE8WA2 z$_h_p3-wlTiQx-M%}pogki%norK1}65U;OJWZOjnvS43j#+!bV>yuR#KkAXcb!nCA zQ_mmCU(bIgNv*CuguHfHbY8g!wPeV6^cG&16QO$RN6c#JbS*4T{++vWpoaI@{LoF zlbMB;seK6hzzQk6yWNrge$DfAhb5j`dt5;^-}4xDFpQHCEs>wzGQWZYc@()kWO-0S z)SU$t?KkdxGJHoYDj=^)%O`m!`5n|Pb6uS_++pM3iPKJtldq?{u9u3p?!BWfzUpB* zBb31rK|PsrD7Z%fID6}EmjCsZ2rYD(c6;*H+J5xeDd?NiWO@*5C+Viixn739wfoyL zQ$z3f{tIG=k!Nic1+3Fnp0GRsm=drpg?ew|hF!kT{3+0o)89_?YjwZuja)3m7`s>; zYv=?;U}>(wKg$W`4wr$+$JpZ&tWRT=71f@qwv0xHv2NzhezDtF*fdN06<9PASP}|| z=v+4ET$2H=YBE<0Roc)AU}A`(1$Gwxc2gvw4=3L{@W@Tq8*yWMKz+QTyx&_7ukFsj zjlLM$NbCDA9*AQrCUeQ@woDl8BKG!hGZC?$MtO)P4c?+Bg~eXhFrBW(Vcc?)V=o)u z!s322_@>@}BWHpr=HN}$S9hW}_Mjzt4{i$ro_+7!`P;YpJ+GKFnmrFMiHBw!iED?8sPr0PFbLX@ULBeD2e; z8K(E$Jg5eP13f<8^lZUghmSdoyT5e95V%D$4m^fKr$&User&?xLqcc48hWSyB47ZR zHXbqr7F$t#sW*=nF=#)?Rnv%K#f#I4Da?#2CpmjKqs%pI`11MdhnSTXyE7xrnz2~w zkub6~XM(=6w_F@w@_M%PpzVw(c}@97f)-h*q-A_0v3IaRdk2?-a997);1bqd=MSis zipdw5`i(EEz%B`lP{mNAy!^w8tEwM0t1@ zX(DD9wbG7`+y4haJM(Ac}5id<7O<)=ugjeKFywLnNc3g9f*9`}cY% zy&U3ulLTPV-`VpM<2IZHK`;(3#5pXSa=FCGu*r^=jP*LJhzZY)FSc662oI~Ym!(5( zy!o9WWXFPwNwkoC*Lx?&$02LhyH5r&FPZTg+`Unwj7`Bc+b-%ZeCI!_1O!6V(Z#F^ zg%v!PcrPCK$#cbbQx7TFq$NNf82em`Pxjh^GpEL&o$o%h@U5r+!jbFLocSIW>73ce`<;LI zEC$Bnocg5Ta+5;M=^^|c6h^2_YOZ0T*ok9`>E}aL*KDfFp+k)aR%duR>au9d2d|MA zHU9d41>H=%lwq!Z*I6s|2jDS?k+76*?P&$m98I^Mb*d?_jxK@V^Nu?bT$hZ`*y5tlyxXdvkE4 z<6pL6f8+-`ulXh*$0~Nmvs?nC2*MKwFK#);-Q7wTMHTT*{{KKUI0;s)0K1q2O^42% z21^D__S$^?^Z{o0SOF2kd^qpRv!Y@-(PlJ(n#14`E(MZ9R8vEr?e(xz<+FFrz(bx> zxo0Q;=zH?7f2ur1?h9|e?BMhUso*m$@2bdo%=hT;JKs~pk{o{9?b-}XZHC@W^s!3a zeV9~x?2G+*l`iB-NA=tm$_e{Rk0@_S|LRdh=KWBK$a(KAE%r3Mq1Ysl(&+vS=U6qt z`*=zwim>Lz9-Nd216^d87>`KlG@T74y~>?m_a)eFGdPri*Pp%0E{{QnTr|#GPfDPd z+xlsRi-tG$Mi?mIhMA}xeCp?TCl&{QX@G9#&l=nC$geRYFyi5Ve)bzjJE!~s zos!*tUNSV=;Np>?lG71BHv^lz$N-{aYxF*;N|!pl4OQ3{7g`cgp}K@uk=!A0#Y?Pv>E6*Ie6iAB|gjGr3q^hHG&QHhj6r#YV|(4=k+nptGz4&?j? z5&yE7a_p2w5c@dC%BE?vx|*O_9M^VPv_ehJE@*T~_Kn@Hr;L^<&sL)|WpWm8erKn7 zE7p|Wr^44_L|E+57xDpQd5qKTK_4I^VY0O*q02NG$)n^xSo(3T4yA6+&x}8@L8vgk z*FXPp{}-`ze-zX+#ZuKUM1hU!P*tL8V>Ku zOgp#33tIWI_mZ+L&07w!ka6}-tT9e3<_c|UcyF@^>wK!Wly0$3hQxchAAq21`3=q+ZO{ZtD)eA+iY%75+ z>&Yt$P~rg7^uq)W%}q4>SORE^cK%D}(+iya*m(ZNBPkVC>-47Sr!%Ew`3<}V-CFj_ z7zv~O>qsa`g;7riv{8HJMX*qAuW!JL4g!-672P!#*X^m2H*XiZ2aY>P!^ygWrbpW% zceWxUG3DtI^chwh?x{#jaW=vrkK+Njz^{_|mk!{_n31?`?}It!*6npGOZPitp>WCC zG6SkVQD4vNhLyhPoOt7=a4J;82pAJNwg7fYJl|wPRyS%2g9P@Le=3SEl3o5wlmnPR zDtC^vRp!2(c|DF=F4H4V*lD_OmS20ak(v5l-Q#T*Kb1igFBfa_-KP}Qgv%n+&f)L6 z(2PEULKb(Y+Bo@`IN!{>aP3*sfjxq#xwM>9a@(akNaWU_Q{~?zbz`n{eXT^!4vAm`TK@n9efb2fqh}<0!f+(q^SAc# zYYX1+f?7Kp+5wimRF-5hQo+gHkTb4CfPEYai_w5Z8+o=Q*>h@<9($Ye@V*15fPXH? zm)=V6fVW}fweJ{8vNUzbX5Qv-|3u9xRFJxgx%wIksiXh!ve>I+{H3 zv9n(VXdnh31M~^S{dWoT%R~~i&;@U(w>YYz(F3@>q*jL1i>h=gA{h?or&hFP^N!P) zc-#_yw0!-q@@Sfp4A4`OMQ{}ayz;JD4mo@Tn9Dz$?P(rMFXp1gWMbdte_j(KDR=C_1L zD~!*q8~a~QdCn1+xSGVHY~+z!7;y`8E4@r(z4hQ!=Z{EH0^dHKfBW?I830<}W-xer zVstTNDO~+59aRNH1!mZ#S8oT+%`QZ}E-=T|3cm<1Y2hALDWZ39#_r4jqT^W67%m%3 zK9wEYlAgs;Wg4;XLJGZ4fQGJCP&u;eY<%iva`lCZq`LJ~Dm_*H$t8-{r7hJxwDj($ z0D&&LYUe*h{Ji7;;IC{4)D`J43LfauV&anL5*|!{oRA1B$IjVdU*Z9DPb)y038kvp z_SMJ<*9K_HP&M3=K0*dj%Tc-GdN<>fvm)EJL!TN0hNxiVcIq$g1;Yg0@YzLG6?o{W zN`nURqYD>tJAGT5u>(Hd`ZU{khQhEXB~01-Hxrhj(Zv!G zyLs&P&a-AP9i6IS9GMHEXv}c>{H)nE+Ib5^eoms8+lXgRd{Ed` zv8YmHJ)C&&9}Zo>EEdsuxDT~oE^Vm$Qu>&`XHili_)>aZWx%pnSGC51au&8Q%EqSMQ-Q~1< z@Ax?b4Xc4J{Gj+ADp9#Ow9do$J2gywe8WSNs-|5We@dhGa&F}2D(wOuT9`>sG`aw7 zr`Mf2BuW;z4&;MgkoopzVD5ibcLlJWlNrn zFI=j%C?xIG_><^d;=hfA4iN%UtM&4ZzJ8>-W1{FfR>Q*wO(+~w*u3%Rx2AQ>&62$y zw0^ZhOj{-1673^?H zoi}1qeO1l+ zN?bzZPiLfoMi2_Tc=Yf#J`vw-Po>h-h`n8Ng-spXE&7Z{nBj_!aYVBgVqArK{0woP z5oL9GQ^pr2Ah*?U(gwM4U| z_+X=6ycg_EFd&mL1;eiU0zDQ_eTXcbi^i3d_HT~)-LTxnF5i1~FokOxKij_)FhQNL zMiGa;=*E3^u`)*7S)9@Jj?wcs^ALDE%diNwR8! za}DUoXZ>Uyz@b@p@5m|6<$63Dq?KtBz=tU=?92L%abq@Ikz-WMs*^}|Z36-#1@m;v^9DGV)n zF_6KE%FFt3AGay2Lb1dGiFk_>Llv)6{Jn}>;$?ac zT;IXW%AS4Bp(1n5V}B5yAb|ML2?^Cx9YljO{IO^(dSKD$QT8PIldGsE!R%sFj;TOv zwZ3Ii5#u4%p-9ctF3io6CPK%_Z4hZy7$?{Tnqw-CK#YIAi5C;}-BAC{Q}AaxqH-Ez z2H&#KMDpW5Ao}_bVTnNB4+h*1Z9b|x<;$?XzxI&)jk5FYj5ROoB#?uPeEaS+-jXmp^ezn$T|bF8La(d{3!@VODhR{~v>l`9S97H5)j%lP_>Y8nw|y^HmRaEakL zxkXp7)e5}2Ww+C2@1{yoRCUcFQMbCm<>7q%N8h9dF1l7?N9pem{>zK6@PPFO^^{c3 zLgvgC?xp4n*H6kN%eJI7vlDPt@~jt%bbC|@jC2oZA7MhxNX2CD_}M`{ zl!F$yJ^|Lalyb3?3jQ1mE1e?><3Q~xjnkw}=XFK<-PHUC4rt23ZCvPv$0H2{7y6gE z*$!LY4g*&CN(P{o^RAm8Ggh_}Y9^AbQ;8g zPat{1#B1T{@nP7_3nvo_j8?hXFdI#l3WJRI>Q-{3c~d1Ce8x@`?p$4eMr))ZfJ2df z!Du4lMX98F!k~jLaGeP?m$(CkW~Lhxf&?FvkqNs<<6%u-&J4J>;L-OgP`9TxxMc7c{T{T#yF4gq{k!PBm==2 z90JjA%0F`9JDT}W%6z4)({+RKvFMgOO8**>jH#L61HqBrWKG@?zm^*wn-p1GjkbNl zjMmRSeb3!Kx`$YFe|KS1FoOqssf85X{(x56WPsMto?T-2xM@O_++)!$Jp>U3-W$fV zRvlcISt2ndI}WS8m3!bmTqcM9?y{?`&J>O)9 zeY%9i-H;NOkCcRbO9EtwL1p0fM$@&K2$U?^Mctw;bnd&(fI>kdY4H|##Y!nu!t5~We*3>6qB^n5%-H0A@&7Z?-q<$Rfx_*=&P z$1N^t`qtGDOYhRkK5C{}Ieb`cuz=jl7oGOI^ZL*4jQxQnRbGi! z9$>|9pjyAC&PlT$U}O5zz$7X3@%>DY_8tEF3XMGL2Cq{q}9!0 zup7Q5PLpfxv~>4QWPmT$dITKZZ1#S5aA*dyp8s7703RUM*xgeoBHS~ltU#lowFxUf zim=BIymDr=GeT z9_96u0aDb7Y;$ITAL9qRnX*66u~au3MgbGaC|zoQtl}wh6;B;fQeR)p&B6lHJ*oQL z?VxcQ=VyIY=B;264phdjCnxBk=%l0lrgE4rb_iwJj->wT*X_S88u`>XPi8j$MRbc2 zQ&D*Z9j)g`3dMhj3YY{OA7DSi!qB;uw*C<;x{x^#%_LH@UDWAPO11c*Ngwm}UO7NN zJczX2QR+hV{}+Y*MvWxpbyfXj-v^pDU#$yn+O%+emO$Z;mY`NJ84Tnw5&j6=1J4S(NVA9x(U0u$QmBI`Y%E zp)S7BG^*dS-@UcE@LO(%@DldHZ}8vX?p6PpmeHvEznV}16CnkG>GJA{(AjB7!&7^p z_^Tj5Wz)FwG7CVb+~ARVPrm^6n_+&WU<6t!)n@kjEc@#&03(y2bE;8hdxNFTjSF0E z$M^kb>UfBvys#EXoR|^5$We8Da9l0QVA}n&6`-K}O*`J$A|YV?jACfKnPFvd=0tRb zb8Q-BhzobVo4vAXbY7b(W6u;_{ZO~d!Zf&Fs-r*EV7}}q9W>IZ7$eyj&?6nyQezVu zREFZ3!}Z(GAlukak8wTBL$YQiQv-x@0>v?ugyCv*t^Pkj+hbtJ;BqtDr{Qo7yvwFB zy1$G$76c2NgHrP+NUI4j0mL=KgUXpw^pL)9j>)qv(dMq+r7i54O8-IJVYdf{yc<2Y z59)SXI3%UGS#NxVAL@i&Y+dZj<%b)lXI<>we2-f_Ei%wdL85+jlw)zGVFdHFsO90{ zul5qs$q2`h1|2`*U#==TgUU>7;(OJ;#t*!RFk4PR|GiJP;*6oWTelyjP|+3;(=Q7q z)NFn9=^q%KGFf+|Nd@cYeVs{W!foTBv@nmbVP6MN89rdp`YHI^i?9FE3^?vPxv`xN z?#If7Oe(cO>4V!vdn>qNs7i39B15!7Qrt_AF_Ts76f##PUq&(4j#FKwSyt4iQ8Sz- zLTg@1GtV8+Fne#?objW2&6@0^SDNN1=U(`pWA(MT(ReRb{}*@x;B^^f70|@?Of1(y z6HUqIPLu5N`>ZoI)M;`Qchw&(9*A+`e>dKPrccq$-*2_%tV$Lg5l<$XLP&G;Naqa) zD{kec+k&son8McPK}oUGCF|PON1+^v#L6nwl@?tOZWEI6OVCGKufpQHeF+Qy7sn%3FHgwCQH_$cXiPz(f1Os zoQlscA_GL&&Q{Rd-O)8&_kZhQK$uH#0^CV*vl(xt`}t4P+V4aeb4O50sc(q2_iMu`ZcdS4OT$soy_4HtJY$aF%NpMbl^#x0Kj;`fJU|aw za68ye_#?mp)iN2mi(oWv+XJMrLq=`~VaL(i zYxm>2(R_~BmFK+Hyo{OvV!&aK(GtIg@x?Ei#Qn(c`Sc*e_;NPX;7=kB%*Oz=%yPr* zqx208>;Nw-*6kCgE4S&Q`+ZvT&V%wHcYs(5@W=XUL_8doifpM~xuN)9&_RRTzrT!M zN15A7^1r>uN>9iE%tfM+`Ik@mRFc)Ca-+cVBNZtRA`#h zsP{M85aNL!QeoJWKTj*b5WV0)RjIE`So{1fYo)P<5*@#bF$A4Rq;ds|dsz%o6V;)F zd5hE8jdE#-4`xDaAo_mJhOS42(-4t-Sb^DBId)x_PR$dYoilM2#jAT<1#w|XN z`ud)`7vyGN%(A8R?zUbvtwZ4VjGF2uE8GLtD(k(YInyKI59+;>CRa)|r>EuqEWv1? zmNb3Z4+YQfB~wn>X$ogpNxGN$c3E%=KlJOld2ri3maa(U$br_XY^$R1?cG^19WvIL zQIYg-snChahw@ci&@YuOqh3)5Te=^jSiLK=Gcq79!CU- zqF7k0_7>V%iKyezyP3)IIHdQcfYtd*FN1rdc6?jbIZKS%mT}cr3eWAj_LELlF;7t? zof0o(J=4x6biX;NFl8vNWcQkj{$~jwzFvu=;k?psAyYsPruFxyxQ{W^epFDP+Px^;9b`UpQGsDx1GSq?`xvh4GV3N84+x zL;*GAk3LIcAA0*)&9=#b@`f$We#2yKjjbe-eTKSqWUey9@4h?;dXp#F-;+7vV)6(+ zx2JA$fF4jIDo*k}u76Nx|M#FElDI+U3AgtXJ@mLTZ2i1@_mV_zchB6I;_nro3ysbQ zm(k3wZD{wESSd_rHtb)QM5x{X>RJ_cJyXiCF;$X;wzoJ@t%1;D_1V|FXbrQZ+G5lG zpx3?Cn@{-F+rvMWX2t)3AvC*4-m|-P-{P7pV}J%nK;cih>!R4%N%*20)Olp1h$(<~ z62P+bdZn6KNMccqr=PUGju+EQc2`$3^w7o-vrvY7dVXvqi-?_}QM!a(8>+AAJ1G5M zV++`YPqf@xs!#bUW$qtHJyD6Vrz%!Co(Tsm5X%?QQU&GM^_(%PS=T_+5M4GTU6k%E z!ePcLSnhlZ&vFa8Ef4RNd$nfQoZ*bQd=<$}NrVzmfhK0mOZS48PFrApo}2Eq5HSk> zYLF$*cys&L2eApnI{)d{0bpS43N1Zzr_{L@coh`|1bmw4>;9vkisi+bs!omUV=I{k zEZ7p3M{~tXK`khYs@k9P|=T4EbEHvo(c+aJ>7cYYmo@_@v5nBo|Gt;U@ z5)QnR*5JZ&Kn^=U(KN+#;WO-X6M*aw_cAELFs|_TuX=Du+CPhG8ZV?_y|zRx6anV&4eU2Qk0q6FAons&4s>Tw z(AcLH-{TmUC!wfPJK@Ld!$sZe@)5@K&u{|yT9tW(q^i!x8*TA)k_?V}2pyBFP+`>S z=nv1`7l|So^LEH}S>yB;?~M|C{ajkl|5x#OJt8nh+Z+sg2`OyKHf{5lTip;cRCN$8 z(`iBJgS?dBwR}waj87w-C+Og9D9qJCNRND$sTM<5ga4GaV1Lx`8STosz5;)bRmc`^ zsUlVID+sj;HwDcjrOi0HXy!86hWf+m1u^tvDFq#Lk(1anth+0OeB6EMH9OpngM@Ic zAzr^=OvmW~o_iz6fcocwxFB_Kmnn<9iNk@y+pU|^nzEbH5#*)0$w(kSl?~& zz{bIGF@C!=gJ$G>{?b-;#cZ?veW&4~id4&zBKE2t{U_hjk!;ZRPFCh_O#3j!m+Nqr zH+-C!J_B+O#0e`Od^K6gHazU2m0KvbQaS|Djs39b;{8vDtk%$k!IABW)>@6Usg@P= z?uhGt)b9;N{sH=%ag3eLU~ROJU{^}WMDhIW~I%{1tuTS!Q{Voo}= zd)5WXa*7mlK7%_PgHCkPe%#J0WDfU`8H#_L0XZChfH-mJZsJv#*9?UTI?HdRS=m6P z89f{ZPZW(!akuQW<^MnxLP})*p*%{gx)vkSv9vjCn{}ygUuG!fRb6*@w21^seJdTvlTfhYWd6quc!^*VLK+dh^6j+@|AFFxk_dgznnBAHP zY5zdW?Om16+66!PqJc84PJ%PIbQU&U?A=QjqQoS5!ouLDtJkS%HiSGN7qWh=MNlr3SEot^#p9X>x@cf zjshcF{NGj<1(1U*P2x%qd3s#p!nP!xK$A0|N;BTm=2*W;(A3%6+FS1%jBPXJyhv;x z{b4_2CZzr`IE~N-JX40|75MZVbR)rp3URh1jKAE=*lk>lG`NjpsvxBt zcmjNDmgV+9JloyTVZ2*pvwJhr=X@EnWtCL4EAeIwwR7(u`HJ)M@OM zJQtPg%I^F|mo-*>(HP?OlUrQ!5Ou--(Jtu;6M;G#i#M9Yu~bugNFV>u{a|d)Bed9t z_bys~;Q@+l&_P_hx^{4m!Vj^u_7XY1Sy43J0z|z+_Q=OH@Kar@6oXlBXf5)vhZg9? zPG;fA@bKKpvEhk>fB?VFeg#|a0oo9YJd@L?VZKA_N5So^9v$&qRxa*Un@(=E z{ z)z4u!PNw{QsgyWqXJ_Y_5YzQ-o;5`b6q;1m17=Ln1ObSef?xB*$`0K7OJ(NQsZZjg zKQ)js^Xs8S_ONM>W}M8CgcLyx71Rsx7Q4@nY_+qjY+cXiA(TM>%)+!hhfBdS&U!9m zym%1>vt>leuDw3!DudGJGF7uHI^HPxhg68tI-`2-@=ZLNcZLCfa(;k5Jw~98lW-5( zsX@NIgtevpK!>35H1wi~>Cbykg=?UT`GdZ*T5V>IPspKkeI=CwH9T9$6k+O@fl;9P z&WtWy4owX5EeKUg0jM?l_vBy%3E)W#l5Y->t04RYR=;cxk!f@gu}#aC8!2)Ty{0$k z3J|sLR;+?rT`5#!^hPN$olC0hVK_teR!Y9lhDxjT;#SI-kL(YHk$)l9Bb66@bCnDiJ4)0$v&eObFsq)M^J%biNNOg9=Ct+5}T)PIkMTk4Bqnu z+GLt93kG)e2zl=Q#jR+QAPriQ_lFJeHO$vnc&ew3edr2m(~bwu@@H%~*ZA?8Yv1PT zD#6u$IKi`4499&-{cWSw_jU(*;)8oZN!9~aNt-RQ=-3~-aMfF>EYb;G?TAUj(l$_{p|i8xeZ?V#^+U6hO=8G`drt^+ zVMFDKLVISQ=Rfi@pb1zyMq=ElpR8*SAhQMvS|c%6Y!B6zsf6ZnANwk~sKPn)DkIS} zw<;xX3`@_@1Hu307Q`OAY{s3eb2o&CB4b_SR$H2rjxQKQ zCi2MP?Y|zHI;8XI&#VK{K@$Y_JE!p2=Gp6*-U7Ym!6LGq!MWcj)YTL0iyzk?`b! zk(#I110_~I=TL#d; zWAP#Upx@9W=Xf8c01I2nO1-h9^al+;LH=PXwPnkE~Dfa-}e7+>sH=L%4!h^qui{@*@ASOQLZhvqk9n$ny~nvT=K7(%xc zHs{L0u|mr_#$X`IB0S?cRiWwX92t}9uv9gZjM;R5P~dYg!@-ez6e#6yhy4LA?X z)Hr?fJs`-#Ac%9)j0T^Z%?dF>B09`y=~A314HsCJLJSVKI`21x%X^m2Y;BVEehY}D z0uch8jBBysim+T)d6bQe$Sk+o0cD0H>6{1DuCW=-zi}JbTxmq_mD<5MxNFzm%%JgFM2`rwVdF#5E-(6#@-mFey9DS zs}WPuPD%q8XAyjuj?uY@sjGn*vV<{^jg3I@v?7v0haRdyE99Gu5Pc}^Wb-q6C$*aX zTXOs8_}=4>me*hWjD{rZrRMatDCuH=|^-R;yxIV zH7jw#VxW6OeigUmfJy%TzEM@={#7Gb?C$|bnj~zXOlKhww%Z-p*-t{hacU6Wt=W2m zdk>vPn$g)nP6SU3d~4k?W&Ec(;><$LiS#Uu7#J030Si9(Pw2UJZ(|~0<=j?h1!aI* z`GN>5bER-ZMmKK3Th0B)ja{?nq$*wosw8#Qb=f3r=<4kBpOgz4+2M>(bziJHc_>YX%p~kH@7qxKb32zU;SwLh)1soVH81~zuWP?+hieMs zL6m?0hBiYor$Nc%N5*POT1uAT{?j*j@2q*=h+pueQ}co!bQFxc-euSr^{=n$3PtVRqqEVuR=A zMF`mb#F?`EhzkZ2e0#AX^3?@s!=D;dLlrrCA3{n4va>TY zv6lzz%9xOVN`z(tJGS!phVQ^ZO;ZB-F;*aO!pv1{TOog{t+?buc~Bp`aHiiU3<(hY1aEgwpUvlSq#j33=Fs z2j5yljF0S14$3!}mHi?-XZ+pVnA?+WUaqY$ z!wF61bv}1;z8#3Rf4?7J@Z7e;udVt+Jte)Xn~wj1aroL0p?%RIMSCX26PG}&KP9T` zOty|3A9ib`dyqD_1m(H}iN!*KHrhA%VeEGT4186|m|sYdcrVH;k8Wj_mmBB#)H6dB zcZilrt<^df0TCas6D4-xB9ay2rV^GJ^ge*q&kCjjoV&l8w^+PXdS_CfMZax9mc>Pw zBVGgIXhrFji%2VTQ;3mYH)cDMC6^zcwlsrc(U-3Viy?hsI~Fp@0+TqoYif3R{#;Vr zeTo}pjIJPXdSlugM&bT{3Y29AH-Oz`aTT8 zNx)deK*vjb4i|9ot?}32;|HhF&KLHxV`0ulUC-0lb5E?V_R&kp#*dUIzulHSI&L^% zriz8NI85q}_FpS4A80cQBzRiQr@c8(Z1-%uRnXY24ASpz{Xeq4IxMQT>mG)oyF*Gk zMd=byxI2b_j@0o@4C)){)3r0=f3Z~_TFo) z9V8$|C9C@0{MG4>qR%7PG4wyo%Kxd5!vL;vBSSvZrevv5lpP&L(zg{P?%ToTyAaLp z>Ace;$LSnK%mA14s)wIsDs*X88_B;MvnMA(Vvh&^M(e-N9UB-pKs%vmVKX}N??fZ% z#DQ^@Z$5Zm@=Bu#Do!S|o1Tt~GWV!tU?JERLd|`HhB%YAwc|3o7X~Cki{meJq{raN zxKF>?#7VCr|BTLcYBZ^fJ$cKX^FSd6BHs{)V#CUyD=zZ@i;a?$PSifeJoqI+% z=xA{o;9=RAa2`wez!N-UPz)0v+3;DrbmkQ~>aI}?X7ISr?|%Q20H6ReBvI=&jmNd- z){A`|&h^3S-$pUJ5wInRA0OqgBu@&;i6KIu8p-QXm9^iEEL~Kg)sgY)Ae*=^>PBRXKnpriZ-4C$y=70Qb* zECAokxVA@$M<0#7(JqkOwhGJLE3q+rCuX<6TiK!r;*T#4ZLM6_FFr~cwWI`hA`u^T z@CW=iB>V~g6Tb_nvCy7q)9$-7PMlrYrGr;xeS}wU(m}}Qjo3^= z@-AhGo}(Xdk*A&m1;7M>LjsZZ3G&AKs_S3h{HRwyIjCnGeVp;(#@sS6g;Nx@Hisn# zO1_>46foxg$HjBYwlU;WE{7AY%DoKkDEvPM?17TlQ;frL?(RC{L;Oxg)9I5F(wary z(|kEzcG7DVov9O%XN%kvnp=4>e`ON~Z`V7$yPxGn96IExcQv6+rJdnTv~=7tz0;cG znaW)GHm|)(SfaChfNdE=IL%on()m#7lubAQYcarf>i@rBIW{aODQIthq<;C+(|A$0 zTLVj=^hjeUC%2@o09PYlB?Ae)lLuY9V+r?VV*JTAfkx3{A$c`f9g~WI+WGhAG7`IZ zv@EKu^b~7atmJa~Lup1##9r}SV@~#Wi6uljh3{w9DKIxrioJCQT`pB+2nN&D-I&q^ z{Ku{|XU#J!VNkNO{M={gG5gN5tlrz9=MO8o1{;L=XpZ-f0`k$v| zO!fP8JMNrFbXFX=U6oZ;74T6ZKsIc}XW|%^_GsWr*+!Q_-JCKoluautoDTiy%Nk}P zPP(3)w^2`$gMFJWr##!r?ZnRfPO?gsoqX>)-e$yS4b={>F(zY;2N?s4c<5AX|1N;S zNH@C`WBX{MweV5t@mg5LJB+q6&v%6N3!KIWv_fmSfsWUe(m8TAtnKhO^?LKuf6jjo z`kF0%6%XU7{Qd7JEoVprmY9k$83qKTDTlZ_qmCWm_Tm^Ol^+){ozQ>Z{-0$aR@l>B zL~DV!CX6YpIeUOzuv=zlGD^2ke&pc=r=r4Jbu+?^d4=hs=sA5c_MM~6_e;C5N4OnJ znoZ^h2VMb|anm@0b@R;U|J4Hi=L0r=Q#Ik+KBT7?j&r~$^j+>M<#ZFRloOj4)Z$?X zjDF8*X{k#N3HImgZ2{O?&Ojl_w`48!?Vt^Gm|RF_e14Ce*(hDf8sX9+N{EBl+bLJ5Jl6XoQ%^l=Q}R8OdYZ%Zfe_?=MBh@0(2Y4uK*sNuKog51#bJPuep zr0e3!srJCDLU*(+`_IWXcBOLQP8BSJ2b_3|tF2g^YI)6_5 zh2P4fdq2y|Y!{%gqM@~Bi8GWXtDxFf29VXpgJM*={vj{IZI{Wo_`uJF2h;C1FR7f7 zGkX32xvUd>uFpbDlvHvA(;lNkfyvM$>N;Et@52Da?W+Sy=q^LCCPV?E6`Z{s7;RAP z9Ke**BW^9ZxU`~PYHuVeE4yci+jrV%+g)#jfd5!`bbe<^0ZN){{8POV^TL^EXsCXK z`>b;SKB@9id^EBYD+x3{W%=g{Q56^Ci+*V9_h}0p`dFbCsW&gs*_Pd+JGa0>1d!Wx z^?7&cKv**Ur#4bJcYZ^gADWkV)|M*=-91=agO;@!T8G_~EwhE92~ zDLFBtKZDa-L-YT|rM!iX9QZH^0VgRhoK~G8ka+CR5)5_oKBKXnEt5&%Dt=(-!M#c9hB=kAiz?mFvl6%!ykO4| z-I-rc`{zTXgHYtI^D>u5fDS?`;*J~>)H>} ztOR$Xz_O|g-@j;ue^9%cG;AA|ZheADy_VHJ4Nsq-%NkY0ADG=fnCCc!_%qrz4&v-# z_p=nQ_%;=EXJGe5E=DKu5O_oP#${up*-j_orPqrwql4(K>XCR6eO3;S6m@QdP>p?k zIGtpV3X|L`q1trvXVr<*k!ykhYb#5J6hw;zNjq!sP2o04Nju|Mv>+f; zKe<3~WVz%kV-mCxXk8>oUej2b5OkJHxuV;Q!aroql~y2dfo8v)Fj@T0|Hg%3?@ zPn(t`BKRm|hRe{{KFY1a-tM<|$h^*62^7RkFaR0+VvN>$H&p1)Q+punQ_6chNjs)? zycRi#C2o*fUsr4chYwD4etOnoJ`2P6$HbuPvt4I*sGFR)%wVz+c82y0r+hu2A3b|76RT7enzvFgywA7)->seo*ilTf}l75>RT zs_L7H3ZNJZ`(+DF-nzBFtD5cl5DLa=L|DdG*}n=RaH#jc7wl8+@q^e-ntUtbjdoTg z_9VxQOUblEdT+w4pCj&f)1(TI^fCWs(=jzO$dfA?)8zOWIZxr(GFDvi`HuJ3(sFAi zklEQvSIZo0v;(S}f;q;Qv=On@>}-uTw|OuUvxr(h|EIluP%`+LzwY8=@&P%)MXYfK zl8JlPhCrvmSw;{Pj5t+{c{#XXM2FWxg#RJ+12XK!ohl1R!wXJ7H!yT&+!2+|&`Hj4 zc-s$oYWiNnPyRO^g=G%p05Ata_8o)w8T0cD{M*R9W`~8-0gFm1vu{PA{ppIQQb}(6 zZg>>#?k#~|Gu<3}?Bk?i6j&q9%hJ+&kIU%M8#2q`e|RhWsTj5VhPj+aIc0r*rOll<+vfPR8Hb30Gjw$^bp=zW&R96jcwbU7~zn@(zg9R5gl7l}kKo*jQV6FXG zsl|B3nV^f8H<+Vyrf^Vwq}`4EYnT4c0EAHv_^}cxQqQ#=3Fz;&_D9mC6shyU^K4B+ zy2>}6@s#3nx<7x+f$GNYbj8Ky{b|2hIKWYoIA+hrT8Le>|4U5qJCJufz%cFh76q(9 z=5?aCULDv0h9Qra_s`ca@m7~4P3-SVyBFxcriIcl_3;OthPx6cKfEkjFt@D^y9 zJ+0<|KKH;5p?Q72ve1I$j_HG|+59(e#{)4lA-`@gR{Avs!S4RjaSl}V3;M*c(CL2K zTr%wY!@0%hg^d3k227zXO$)k~Pr>m$ATf!dlpKwt4HNKNIEklb&~TJfaT0W*e*RPT z1DIykA@vo4z%xFn#R@!JnSVAq|C67ipuj8Q&p&AyA5+2Dd8Wi!X0MBP*u_?!0iN-@ zsq#k5CMbrpUXk7@EAHJccRO;xm;InDO)iSO7*)b>v8$@DD*Pvb>`yp25Qpt9*=2Np z<0DoO8!ZuBt!$Wq)!3Ql>^8gzUh^Yek8F7}C89$jvPm)F5Gpy&EUA&*rV$q6Gs`6> zEh(Qr%B)?x&(LOBmE~Y9IK@&aEC?lQwl^ZT)=0< zVI4Sug@U7dZ~X+Z!j?#4=U;j^@9{^zm*8^u>)$7_ws>5yR1LN7*WPZo->j&X`X}+^54HnF!?0aPO;$6emnMP9 zW%lWmOc{63&VrEGB@;GpSVE;iYa0UbqsGi+$JMT+H$y2LcyHf)ety^D>&Bzgiv7ou zK=zm8@~>m$S^Ajm^Y~=O($QcRZoTG7tsZ0YVq2#cfojS!$O4fkqfBV$=nM%oUEL;Y&+L^`4w;i3iILB;;C6 zhSXXxY2)mj?Mfqg7bd(aE0j&a(ka*agrr(Hl=_O!)U5a~pz|lCV5ZHHQy<_M^w}X? zkGJ!f-NM+Dm}QEzqZz><7{oO|TP&L@MAM1%Nb{}=z%0=n{VneQ3xfX-)ttz4r%w3V z!h0pJ_nU?c4ja&93%X@&pb1g*I~DRDr@hDP1bL zAvw>D1lR7CUHKnxcDuMSt)D4orQlQ)!t1rV&;#x1J9Y;_4X1L&(5WB2b>=NZ&mxzx zUx#2wMo=3x*~yf1l-@xNV(JIwd}{*3#pPnN{uPs9F(M>Bs2u=tW#gE4w!?Ev&c86I z7tM@aZ&^NV$qpZWn7lRZ_WtKzg`mP6 z?MQ^}nkZ+FxAQwPdytZ+kL?CT(ERRB6ID|mNx>k)YsJG8&Xh`#4f#}NU}6W^xM5_C z^srjn!Psa`>@u^>wi<0KhzQeSDu_bk?eI0!x3@M~gF+UN%^fyi*f~Nbi3%f@#qZ_QMfy^XyrEvyH z;Nq`T2S8Va*HPG`?SoC=1b+}z$YJ&lD>HEX)S&Bga2$=IUAX3JDr=9liD8y7^Id{NDn@O zqyTy83kvZA=g4^QraGS6FH*-~g$a3LrSRFmT^vn|lQN3ed=d3|OSj`58ISnWHLF(6 z9>NxfzS#2j`Y<7Pv|2{x&E@6RmKqa8Se(VYO+8%AUDd)2lT*oa>3S+xDC}bOz1qC* z4INakB0p`mO^N-%@uo*^wj#4SaJG_rhwA_IIAo~(P`wt{uO`2A%e(h_8kUmY*nAIE zH<mC5FWE=*TL7vK!t3a5|Wfgt?yOKRuGMtS&1 zA;x5$l7sN@?#~c>e@$uTA=>J8r=vJH+FdDn%+$$lQlV6&?PXqa6%kF3H|n`{7Bywz ze&E&W27}SF(FDYW=R)ItEAN}Uh$nu@t3C+WcO_46$%008pog8%>UcqA!>YNQ&n4>a zZWjed*N&=4zNSS$2m6zNtYS&YLvfw3;p`#b_L;ftHWU~uf?Oo!Jc+UlLm}|Ro(HqF z^AEfU1Cx(m*=6q^n%a~)pV3{cu!i$Un=O|;64JukDyw_mKPXY%lVm?3xz#utky8k8 zK2Md_h_BNu4c5oQNivQ{JJeOc*sM0}zv+*}5BB3+^P;Vi)$za0K91*Ie6c|fFE(+; zO(J8s`(s>L1$9`rk)>|M8Cm^-@j0R1YKM%BWQu6u+n13@!iB^}4FwwSG+!ZOZ~l6G zK&?&i4(?AOy}6bEE>FXH{+e?Qonm>@sz|pO zhXQ}&j$NH!K2wG_k2zs>!3BR&hT%{bV#H>UMJnU7$ay}}`0C6`smT&iJGo zlk=yoTd_~GBIJZMaYuZ{rXg}01;e(U%9(wB7>{3Y6ANrivK0vX&MjFBdEGH#;-bUW zrQdxR^~cZF6PhckM_RloJp2y53Ei|BqwbIAR^8Lbh(W8*j4E|s8PD2wHC=YDj6eI` z+(*F6Nlz?wNoOe$P#k~8+M}k#nuwr6&+os3;xv(3uiW(s9)9v71~;^%91nw#wnvqW zj?nemDVXDO;Z+}{X!KbmQD~1uuV+|#boI>k(4fbX{e}=~bmqS!cyCZp)y+2H4h^f` zouHGvYI*wLWv3AF&*#sMmK71YkC=9}=L2Nt91uDwcUMLKQhCG9#xDG@U){yl``e%$ z^zj;QCyV>7?l16IoX5h!#}(dgfa8_LzFNn)Pu7?iLu>L9vY9Rl;j>Q^Nv%riadL3^ zdI0ab$KA`Db!=6?sYI8mIUNAGRFl(1$fY)dpD<#=f`FnHb*Z^8LW|ez@)t?fo~dpD z$0oNJHnGp=bAX8z8&^Z5^D6%(t`0?Qpts`s#Fo>i7VYshF7XakT9%R;5F*3Bxyn0cq*_tBeIpmp0{qB^3#V| z0$VcejMlfCJB53*$5{rs3YM@4DGMz9kn^#!~+JH&~bDO;K@zTHciejyj@rCe~9 z8f$*V8;zZ{xO1xkeB|QNZkRI8MShXHU7Kv#Y0 zMHlkVwZ8eeA;)5f5Lpn90w6nDraO*rIwzO0Og1Pl18!{R3Z$_2&C)>#Nj-s@7qhIc zZlLTD+<6uhllLp{`g!Q8FNa#;2a`7v=>)9bUIZH)1VsdtFy-ZyET|CR0$Eui@=U~m zZ*U5<_ByZdcjOP7@>u?2Nlqa2Lh!m=F; ze17^pw1gP)H+5*aZn84RmC8d+5#?A)$yBVhwOtKiUH;5>%2fFS!|?^SxMnv^=Kg-O zWup!@$(hx05~~pjiJ$-74}qO_ql_<|YjLC2_D4S%A)#$wF5+j2942&lXFiGKkA@vn zW^H?E{RK&IvVks{$JS8QoW3Ujgv8zk0FetxX>v3Njr(n?BJp=me`(3fliB3ZmLyNqMP|1haLha$$l=CWpx zGblCU3D-G*5I9`>SYhKVu8@!0$2pyy0;y)C2s9cy+viytUV8umjStAzV}WjaUQMdD z-#v`CM0Rmk%@zAE`wGR@G*^RZp!rkKtY%VvX$kqBRJ735qZe=cMMJ;DluBDAHr;V3 zvFkNY;k~k6sdgZ&HKScmMM=615pIRWc}e%EK09Zni;Yr678D@~PI z>oA4sm07KxWc?Hi#@H554v&s|&ge>*%*y+en!!k9EuPhJb$Y4T|K{wOh(R%1Eg~AV zPF`d`AxHFeojES8xl0q5#UIoD-n8yg(}7qksdR2yh2h0}$}7qI<$7eHRWAe_)y9Z0 zblCj7cZ^EZIz*Ym;qSu3ReBm<$Rr;V)n!gW-Ot)r>O=&UEr<$I15&naXHZ_Xx;|TD zg=G*`5~zJCte=Eu%bq)C2@dKtzcASr3jYr$4rdkUK?P5S0pCPb*ZhvL_S;cHz1p(% zlS(aP2jkv~_w(Z;9!_NDuVdlA&G-LtfsPqs^Fx%|>EdPg;8e}d3#Uq0$P7X~egMxw zZQqtY!VT}+RqJD(+a5MDZgtJ#85I3EyWGl9kF^qxS@i#Ipr)$Lg-YeGFQKB8;X*Z` z+vh3J!;Lp_&weEI`l4;gQyVCe&fv75z8>-^s9!j*9mPt@kg9D^huuv0{a3*=n_QiF zb_9i1(FW^7b+dvvXop&l4wR+ZEvS@9Hs3Od!$d;0?2)|YFh{jqO_f$KW$vtwj#Rnz zNuE|Pp{V*y&a<@iOIAzs3pw<9!28KHauEw2hs~fKqK- zAMZ4!U(-V#nbBL4|KZmB@9RBbTps#gF91B)loA1{>jsNv+}Z7*ft=y*Wln&Ey@7#zZ zQQAwgn6o0WAT^pmmQNH)F_m_O`Wz`Ghhp)yGFsptyb`_?7&uLihTI|gex!egZ8oJ?*b@wCW;@Y*6o~# zM{|ih!n*Aq_PFd@a5k6(cQK2*!EwLSmXv&cDD?M&`|lM-01$Y4-QZMSknS^2F+A@V zS7nd!xeIp!>&oiL9!&iJPV$3yXwN=uevn1i?p_l zKIZg_%cgtydpD7%=9!a^ftewMfpwj1y^OpGJ{x5s=ynU1km%$J6QQfsb6>X!))Xm? z_lb$;U$$^zQK%R-QyW4VIWEY`LpZtj`hC-L8=a`*TtJBzuFOrIq;(%tI60wSMJh*| zEXZ~2P4;3VZZSCW^yW_q6;wgzBu9gJI|ZQ>*wnk5>aua`O;oSxYH)62pG5yg;L_F! zjpVTVHY4-N2ZQ}32oTTPJocFs#Oe&I?sZ!O2I$DHUBUx8#MNT06HT(Rdt@FV3KD#Ln$m&M@wO2kxs8&R55`Ee!QJh1PItWsLYFL&r)HUqIR zla4T-jcNY0Nteih&qQbatQ>J7yz&Jx3S{s2lrv1{)7vVieOG(Y2w(WQB}l{x@bL?C z&6CENKN>&MK3MMLW^svE0goKAUvt$p$+tBNnX{Rx*(6ijo=q(uxrP_Swv(xVluPWvVM%Q< z$pk*uVcC@NyqNn3-_F9Jp;ypS_J6upzx;mC@}~c8t4{zF7Z- zthev$U0Fr8SCyp$#xBOgf5joP>H%`4(;(yyA^MY8PrWhTQ4-4+pYzV=TO22ii>Ce3 zVPnU(89^+aiIS&IK*+jx$$d!K2Z4@om=|!6hUzxH&26&BP2iy7_6d8Do~?|COh%7! zs*?e6Htid9{#6{yGHT??8w0EZb$3;F_0GW~_^GDE$54hoO68lggBEH0SeO7zunxSK zF<`NJ;(%u_gTOgLbXsN<;l)LBN&w!mvY4cspup^RVs$3$<7VLQG^)Ko|0C^#iA=EY)0P5m#!&>-QHVOQ7KWQt_W4qNQ;fMONWd1eNu@Q$aPXNene_KNLI zVF#*}1~dFcBt7-~iK&c|41Z_lls)P4)r}i_q!g>{a^ncUY{Mf5xLxyoqN84jnk-%% z{rgFpHG|zMeNsswPu>u9nxf$(zG}y%=41l@;`us?1fgaQd!rWy1N|G@bCnZYJi$gU zaGP%L+vX%M_uo3(Hat3?bkQV<#TVr0F-`6&Q6y09Oq$`bD~IM68P702CaM>(z|Fel z-$CxMe4wF1ZeqE-_Bi=q$6KA`Q}05+w0>QQwmNOXtw?(&#GtSOw;D?s9Nofc?nm=Z z@i3(Vx1O6OIrC|IEQ`O^n@g!e=YwgvmUzCkw+`rRTsljqzQY4H353r0`8o$;T497^ zy*P#O>1!pJ1sP6ix9vMnz}n)z71d$LQc!teHA``-lJ)_o+z(}$wZ_@aoqeZ$>Krqa zTX4nv2Jhb~l-N&HiOOT^IiGN{vo^l&V4B1mm`Q)98`%VBj2Mr%N<&dT*LX!`!mh&U-{sB>*)RD8QyG8-oqO*x z&$AiU@$pJfhH02oCevosz1Ab`3TZwNob>E%sR7kRoP(Lm@M?R*-S?eLrWw|()n`;H z^!UX4sPFuOkqCJPO&kWoaz+?8)!wH#*!Hm*sD&85g5hU#7=HeoZ0?`bLctDk=Ap@s$e^Riu7Z7Tn)Fn5Sq;~+0)!Up>4Nw%fyrNK z1P0v5DvY3R^$IXK9(7K_)0jWug2mI`dWku~7x%YdOVhnk4?tuSSpqY+AY|6C1>lvJ zHH*KpL?My9-8yi@qn2FGH7Vn#;4>nxkkV`r12d*uJvhZbdoL^O=hHOnoi z0-4iPzP8&xV9g5M;ii~4m>k`U%66H@204B<#(1+bBRLC4_`3fqP~-K0)+u9Ieen0^ zdF|&Cno~bD5nYNoS|sB~OQy%S_n2{OD(0IVu2N3$6b(q^Sb5+U{QCkze`tV6r@kkV z{vbS&9hcvDXECuOAVa;2bM1{d;Vj{FCDfG?rX~q2e0!$ACo8lYc~I+2t7NwjMDgY0 z#^74uT_5kwulcKR7eCxWrlV1}a~=QQ>j*;3SBcEYa<~emvBQj2ocUkFUZOM8a710# zJ=t!-=UAn@7<|`gV*?~5^s-hbefKj7&)`Qlw&?X^GFqeUr(j}3t4txu0>W|=Hbr}; zNm~=1h^QjYMD;L{+E3Y^qyT%=HATKx(}CbUezu=wBC5HMS-zt;*UgY z&CySJtb+V7pWXH;`PemlkDfRD@+08Et~se)mxmh%mkbuAsO92iCz6_n*oPPHbt}qv zZz>Jg+rj_NiU#R%+ z1vn5re+xtB#2w=n1zl4;0+ubtl~2a<>rqvY(cJf3g|54Ke1Q3r#r7SYoQ~zl6wDh% z3l7CPXsaYpf+{?giQN8^aFA_ZMM%Qs98V?6{ax4!v0|Z_cRo>kiGei5m3^o5gRlQj zJlaDrnuApj?nIg$VrZqPZzI=4u|^ouFa>et{bxUhLr;E+e#Rw)af2UD7A_~L_gOo> zGyYSf1d=|e;NjO71jE?zeEff6doW$i-AkvFjP8i69*Uga^?6CF0EeCRUzQf^iuCSp zr+Ttx8l++8I<1ubTJ%;tl5D6NH{PHE;3U4cLS=t0j8$h!TQ2Es2b*)D{9Uq_?06v! zfvy?ugquye6uT_Vlyyj30^T9p-XQLFrN#*3@btWpL3KF_C-!(|xPE8VIIfw`0;(;`{2 z(a3R^yXk#_E+3xc8MnY8^}B6+0>~?5#ZmMFndItXW8BlStN)U&l87IP zY#kk2B9#|69!hYK)J!-&d?K2L5(iD{DXOuA@w@HQkvL@uvRdZSzF>a(ax(xtA=MtV zFoa&-$?iL+9R>eOam(2-SDt`-+*Z~h1WI-`luS9fUO>1*QYLl&C{a1><~)EH7OT*?SL?B8+@^FHx<|7nC^PXtXUU>8;dt8)3tzud*|yUZ(7xam6)UNlrca zC@ST2`~D@S17%85+k_~SF*8D>>9G9;sCYvKXu)AvVv3GD{^xlA!~we!cbe{R4Y*i$n8pi7JzK#xm1?;J2(8Eo}gVfaJD4 z?vZZc3$jPzxRFlYcO?qZqezUcNGw`h6Lpd#w#oPEbcMErI)_Q)J6rp6xlT(>BDkNi z2FqiAC04n|m2kGrm1tP#jF$&Q35!Pt0A9sga8%jph?CR5Wd>`@&EvfuCYF?=mEFsm zETr3vkpku+O0O3bcnVz)P>JMjsp7JiIlX$nr?z&--ogdcD*sUp$*U<)DliuX?I@wXT17|YZ@OD=s|%*Cf87Ni2M#JMO!-VM^d8C zHcNb{_Zae_(gqG9GgB`1FgrQ+@#=3Mqe(ex7TC*s_2Xl0H}P4gQ4!3B`WuTcG?ET_ zMBDxIdm5ux-Rj-8ZxEJyMRr58Uc^u9Hu1y)>C&IpnpA0$bkAEJbf|OFDP8A7X#gyk z7^y%jG|}nv=l&1fGfZkMT~qT91v7h-7-2)^Gt0CUa#$~8WlY}Ib(!_Bg8+o-MjrEfBB#QX-S&1tNrbbXE4S3^E5e{r{%>Y zRnmh;7^Cr26VdXyZB$%_7n_x@G9o2jgV@L(wtseb^VhydK;w0cUdTIrD>K7-u@B%n zOr|VeYMXP@XWlpaE>aut`ZDmYvU;eMRqIl+r_R7gQg;!H&>_v%zB<_)BtkdI852}+ z7{RW8-GCyOFWEnYr&mF;X{PEN6Y(p;?Fb`Era;7h#XnM_ce24|((0CPKXQ{Uk0EM1 zvBnDbGVn|>dDSKg+9e*xT0J^tdurkU*mr0{^%Z@XY5D;9V9&2j4RXUJd5#r$X{A|R;R~L1y|e88M6-EX_uhccCq<#C%eTYdkr-Fj?|C}FRdcK>|Ks}ME}>K> z?)=~@;zG(TdYUOFb?Zj<3k)5I`;%7pF=Tl@>LbMT%3u;74G>rnpP2n+#qtGPmI3m| zmWQ-MomvF`yZ(l8UFndhk*5@oPOi#a)2>dld-DVGt28-cU=tR=#B}+NbMrf|*UKi6 zi^TnjJcG8J+YrE`WbT|O61a{UH0 z+o$0W1!t{@8dhoP$%rUPyEqz=jhVg1uv`-~A_)Z$;497E4L3leO4!@!Elm=|<37iB z@_f-Cp%%5)(efbu?gV0V<89*IWUDK^s2e$&obId2m{gKU-S7puIg!aEySw2cI979i z7|y3VS`78WBrA|=h=)*tWt5GC2p9I&LL}Y^z4Ye1^gjuOx^s9_MRWKrx#0jNZfxDHxv=&v4n~_&PnCjI;$%cY%>`7)Zw5@alP-dwM+#4^UJ_%} z$I<7EtBem3MLRIb_z`A9g@aFC;KMhXoI9&D#~hZIpV++!tkiop%;NsFdTdZ6I*#`~ znRgRHeIb@8tTG&klx#ECWX-YTw5xT8vBEXg~($lxI zssxeYvi@&s>ldM4=*p&%JhQOr>$)p<+aEN?-=qZ@N&saHEu^xi?&gp^rjH;bUX`w8 zKdpy&b4HkqFC6NTjwOWFLs`{EU?m+`KIo|*tIN3DgqN^GdDo^ldt$d<+`a-cknWbv z=?cq8cJTB8sYqi^D;3T^0o}w_Sr-%QC8g=keF8`mU<7y3c@0{StbAv6-2p1Db<)Xd z3NG~PYGlz@P)7!j%=Gu+gnAQ)D~Yp4@_kZ_gy?^mCs=x;<3mFBZ^MRaNp(( z-$C=Yv(ypHc1Ic=@4YHIj|BDbPv@$ld{{LX{McY;8BjV0VufUsZBp$Uj-p*8VtiE# zo5LVRYhb&5_hkT+(mBuf@t0;e+${C(aaucS%QpM^Pt>c|Yd+Zu-JO$%|7MI)`=CfP z9Pf?Kij_Ba8hvJ58nL?`>9`85*9zqTe@hsPT7xIN0@@OT5%s4@ezKWV%f-vm#vUfZ z*TJZ?bS-Tn$A=W4XzokJxG+xpcLS7m$eBJXn&)iG{BteRyJOAm;S=&eB^y=`z-!TD z42rtf*Dhkyjj6nbYD?CAC0~+}b)vX0stfAM3>GWs@iVBDR(PAwb0K=P^#s}bU2A|Z zMT<*Cs#PAs#Qgp_4=BklETB!AAKw{+rOS7cA>gfQ-PdXD4_7`p-Z)VIvoi2GoP+QH zTc}0l9*67IS>RFm_L+kT4%qW?e0{ZEN^R)?%s<(b-|fAdJ7%ZK_#gE>5U9Ss`sBOb znbwna-|V}?r(e1LtBwdYLKv{~CxFWPBE08|k(67E9-xltF=65B=8dgNer#ISA zjUdpmFQJed)WD5oaMv1$YWWGL%g_n;TK+~iE<1X7q<@tE)*@YjQ{_VD=IOBhxH4|@T?W%ZjvZW9hb zYUG(pK+_*(dW<`c&x%bG$bW)2yv5XR=DQ{$9znjRCAUe$cqFOZqeq+iF2-mie z@Jsb+Xf!XqkTbwfH;eB!U6dsqpm?|2Fo|bfHh<#*^>*QKXy8nBPP`s5xc)!Zn)S4ss^y;R_{&VEsO<&(}R zlOJ5v({XNqD?V%|%7Ly?JRkAnoAWjlVxLkW=)op;{`37=?H_(*nxfeB^R<_}E%jKP zXkG=1PVPF9z`W2PsaU~Qb9AixYuev+tK}rtyQwdw{KNX{!i8_2=~|&|mujHj!05tV zQcnS5(#oMjRlKB%<}PU;l7vG!{NSHZ;2HL zKAE9@mlsm}-2vWyeP?GxS9PrhGg&qgEn#9VGVYmn=d|5g-yT|+DaMYc&r@ks7M1LY z+poLG+Xbt(`1_VeF2TOf8yqWk+@s<}Dvl4)Gn9RMI-$V4;2;a7 z7P;HfK;5IRtlOQSG;D;>)0V5rf7(*R79Y$Tf=2E7d_*oMU*b#!6gJJhqB9+#e@{9XfqiexXPPzo(fdol;SvX&NXzQ)ch{d))aJSf zdM>unp;X=8q4O1q<@MGjlmsL3@+bQIP#%Re+-1)uJxw8&j#k6}wiyRcPpbXZSeN~GPSHd=y?iMmwpk6u4k z14v{S_fwgwHUCMVdab16`K@-W8pOhAa!Hj|A)&u zT)Z2uv&K}exJ)wr-tqfP!2sb8#N&@&(Dc1>IpURJDXqUrvERk7=P@WfuKiyv0=E`| z?Vi`vnadlg^X&JtLe8+{E{^QooikwD3ns-@(1GF9^FPWvMqql+$}ldgQx_VtNkSGZ1L2Frmb9p~73l zBzw7*cvg6zq>@{27gzHWZ4yzd^esyVURvt-zUo+kD)_r`{I4ql6v@=D69ceWF(*kGj z;KMedyFeQe6JDI*?wJu!!6tVsv6F(^_{c13VPp= zUf<7~##TR;X$-!j1?h3Q1rjk&$00npKNH7zf?Rt7TXL0j`6AvWw9_ z+f#Op5q4<8kqAoOW_P(J46;AWK$ck(j>_8q@2>0w(kNYT^ORI`jAs!8phm>29ElKQ z&T7qhc68xYC5r<-%on}mark0yNLySMKd{rgsAv@vAIlMXJ@t3%tt`7MD#xFv3|#QY zm)d@#cnTuVh-G?PP8*+R&r#!*5r<^E^d|oy;Rql&1OF8^?IR%Gr=ryy4eXZVu&&%a za(YZQ$PyN7^9w;ND*n-S<&OUN2;(k*(6qxXX@xkNUJ;b4m6sx*2=%Qm=*f=@5(WG& zO*^+OOB4-vCuGcC)74CloIXyaxtDt)@F2I_GGLw&nYP&$ z9qKeotJx_5M8Hr;m6WH?p4`o-?I8sA`vh9;pK&*SnkJJ?&uoc@7b(d^uKBYB(qwX; zu1;I;eHR&*U*eRo1-tyOz%~;VBuV2U*gE_0(j}U?LntBPOdD_TJ#Z|bhoa|fUk_9j znbSRxS0?`k`skVfUueJc|43(*mIA-B-@iEddI(T_nnyf>LcR8>PkIk}Y%enfNi*1W1#B`Da7hjM~lqia> z+P@z56{@uLrYI~udJNE(z#;tHNhB+R0=wr*ls27DvqraeTb&ttQ=4{ zQAn1$4ApJ{U7cdvKT>N3p{rB;Hq$Kl-yC%6UEDB|jRc65GQ9*`y%5D@V%;zLM~VN( z)>{X~xnsWQqN@9!rs{dlIs5Fr_F8LaGYN6x&9kQyflD=+))o{y>o^U`W5H-OFiRx&NEj&a zvJRXHT!KZW&{2Uz8gBE25cFS*ADz*Q!C_PsTT<(-ds{lX^yLEUlrIu$6}fXqF>m^y zLrwl$>T4=%g}GyJFTAM_wLf#t26P$p>mRkB)sh@JjDnD91Cs&^THq5?(>r-M8}Dbj zCzJ;WSnzgJkOacU2)Vorb_)Tl81MhAIxCnOdvi2EGA;KJn-ZVF!+*!zZHYO|Qp&7B z40H#6T&%6)p>@xbwW|D_Gun^z)FL$_MpQ5@Mz*?*xN-bYGZgdt{NXq}d#___`R}v^ z7aK+KXdWzK&Sqk)Mml4)t=Jp?Tg>iVtE8Qn8E#S?dE>FGyPi_(^WI@D{&18p)Zly0 zWHA~cwv{6vSdbyo^FJr6QvVQbiJlV#1J+|Dv8@@IVqH3aTHgjCkS@^ZAI!yz-nxIa z9R=W>Ec+tjCU)&Gq%`M-IDHNUmTJxINRn<1*o1E-1;%_Esy!=txvMbpCCl|wb#Se? z`pM{T(Kj#TiZuQ!)ybkOh%&jF1RNw?Ahil zQKrvLB`6Yu4I+N}da_KkMfVI+qiX|pAoAW6l!Mp(rbV&Qqf}Z*4e1o#+)<*?Gbsrg z^{PU=XzMfIEPa!luY1RzMRT^;R_U{4@Z3;s;cN*J2Gm-#=jLKka}VeQmHB#ndd_D& zHUZ7}=%Ig6$JRem%J}y8U0^Mzl)M^{-(5nUTmQG$&9RiRZTKXuk*2ha1y=`_5R^jjTzuoYbv-o0QrLjC})3P&p2_`B^vTI_#^T!Sqm*-`=0|?Bb)p^jXYZOuLNPE2h*X9*PdH}H6&f}to9dhLr z``j>vLY%&Ki0!}|Do(f_cfhqg$l;j7sr1!QihBuO26xzp z*w7=Ne`?M*S&>XQyj9;$QsT|OJN)Ck1>UwO-~$c~yTAPH58}P)Xtx4P#kD|**+Ra_uaYEHsmL!%Qm1jB`WLoK*@iOf za`gS4fMF4vMa1LH5Y|iMziR5WOwpdiWrC!PXsaDkymUz<{&0BD3+XH9Go~#ur@P;;sXAo(zXGCKr;d{kCv+HkT9An~ zd4w$MX9=w&$dztikKxWw&mttYnu{D2w16WlmwYBwPi1TJiuAYx<#LlkHC9uO9Lmn> zT<7494BcPh*N05Qb!l`vJD09HwTRRHvrZqjYQFB+GsF7#Cn)teB%hoU;{eJUlA&JF zf1Onm={51}Tv~Dy;@;Fz??-{PF^p0Cxqn@-xbMLVKDVE?D^2Q=WmkO#i@lG*e??S3 zKEz#{^E6Rz6(mff_%W-)RjsT<6zZVPfsSyiIzzVE_Df)%WANWZ(!Z*VzyG5wWoe+& z-vYD4-!-(~55Fop&L+$qpxR(6-ln(@W=#s>n@d~%FiY^vb8^lA{uC2+mQgaV8=D$` z_gSX6z!hsS+!tiSpHj~qsxA|Wo+l^%Mzg|hb9t92hIxcEz2A;MP>CMBoCfR{s-24Y z?1jBh&8+XLn(9%A(_5wG|Df!cdMub%VcxN&pkMO$wagh$Kh% zJJZ@SF?xLeK~$gq!Hs8pl;0cFQBXhZvQ z$CwH<+qG>R2z|7BQf>#N!;iq6|GaVKjGZ( z9}l`J&QCv`qJ`)AA)n6jT1aof%kn!RhDcy>r8!s{wYZZ6C#cBeZnW#S4? ztH^Oc0=A|5jBvH%aqgSf^KXwmsXd{trhX@`;0gCbWXc1V=Yn2_8f?C^x`jL zc{3dFqi(~`A4O{R#B<#4DMu0iy?yxuMZ06@v}Z-TYjs(lAV^wTO8D5rzF`G8fsX^J z!5i~?sx2gk0DReI^@K4`${LrM=@bgj-yU&e^m4$4a*dPUN&WLl5m~^MM8aEM^NaGP zdX_&bt|)Lbv3$%0UbQ=sf(314U1Ex21j@YuVoH33UBkca9eDFC*G(mR32P4Y8A6Ts z1H=smGnbkeyoS&s2W^kJoi+pRGgfzy-g|_s5FgQnduYz(INRQfX317LISMUFA@K)w zR)52auJoun{8plr#uR!{j7r@i^P8mwp zKqWEXSngN&8iUp}705@86T4VGIyAIIKW~i`IT3p+xKxnFY7gsh`ZF9$2+4FX32u1rtwxACX&B=jo8cuE&z9S;63KO}l zCY~_(k~S^w{RfY&bb6sP(p`$23>5wIApzK$WwW`0qX&;Cvc`t1t?yj#8V@h`WK$yH z(9ptP9gcq`LvWt}`0RA;*~Yi{FW75WEM}@qRkEdU{(7>tgWGo(H~)H~FbE2q@aY&N z*WWRmnq9KxDa@;#{DBtY`Xm&8 zx~vYqqbtlSr~2Dq4^Q$m7lw)Ui2uFUX^o%WNx=^_zy9_b-pTpZJJzhp`063eBa}Vq zRVR=pBTgkqdIxRelI$5f5GSdw%SbL)ET!f~v1j*^>P=S;$3;Lc;8)O=Gh#W=woZ}LR~8Qz#skpH zKRZ!{ZeI$_1=cd8R%Ne$gYT*dqaeo5%qJnA0?$a1T!?-Uk2|t5n0NYCKMfArHK(r`*RzKjFe87t(prR_SvLzvE^Z|97g&4W`g%U|_BI3le@@9$U%n-9yJHNQV_sSjrQAX%=|4@B$gDu{%E*)&x%X=l?S93 zAvHD-lb}!pMVlGwT=H2pEnafd9(IlDaj)dw64}-bDRy8!7smjvPlIo_8bTA>oSwm( zwG~4DP4$PLcfV9oZS>5$Jf_>ahPSqmmAetx4odG)PFoJ@?u`0zZg+ul>R9g~ZFa#0 zCktOxmFaL+PWtVM&r0s;z{qC!I6@vq6HFqsAO32z_15xOXtuED74d@|ko3^7d+R@Z z)#sC0kA^d>K5io|$J^j{IQA z;m;64`sLirqgL23dt9G6v!ua*lDZI8aJ|oVG5W_St z>HOFXcz27A>WY-H7gerS`Q;_vJF4AG7GZmN@=D$k1Gcp}U>UhXtmrxJ6+A7tmy+}{ z*fv-Jy*<|Cd)iR{SHS}Reyl`U3J0}y71c)g*;sm9jd<{{e2Z$HNnX#CfZNO3VCMOA zM$IDIJnTi-@;;rrqsd#OTbsmXrdANniOrXs zuC+n4qb*_=~WrRNn5lrLAr*HSn=36ZEqi*$Hi?oc}` zD-5?&*FmLc07BNsxSD}>&+_9Gxeo8 z`=}^xlbG?o`D&w%B!0-8X`xMD{2=4Z{JY>h>GFq-I>mhiqZ)uwPh0iJV0O_2^Dg!7 z1UH>-AGIiJAjbqW2bvJPbJ=sQ91#Jx4D_3pc#-6eC^~`UdVPp^f5QFCjBs6OWf-Thl^0s`@sRaZpo`$ zlk_tYYUD(LZ0jm5LAkg*%@&7kO>s~nw%;%B=>$$0PN%4&Tn;y#d!t2$ zB&{w-k6mz)ff<(6P!bjGO5v8UPTx?@&yHb_el6`$gge0TV>70peVAtYlxVmrK3RSe zz+2*=#JWpPnpT_$r6fys_=cpEVi+!-?bx&Y?CcpZQKmW|gBxBqg^_~~@Z`Fa_e26K)XiI_v?I%k|IPg=d?2u2SMVB=W_}=-#=1We+FFM z?gE{CFouykpZ8;!&ROT#I_uiy_{yA?QgtnhIqO(UH;GrB^R>lbOfK2LZk6{5=MEpY zkd(NV!Io>F)OM6-q7D($NUI^^tbm{1jSMWASO}Fn3j1jqw&*3n{Co?aTlh;8Tj^xq zd*zdOEj!^CWAJuDlhtbr2v9(}6p4}O;#b@UGXtA=Uex%oLCP<=DcF_!St;07mcLes z5Kr^!u!Kp3q%6n_zFyw6HE-ZnDjeC?u(G&G%Pxz2ZIN9{>11eliP!Bf=>ONK2uf3w zb1si_JL3@%65{mYB=wHeb3vKu&iXRC*pIpcHGzQ4kLJwE_g;u^P95=5L*K}Xm}(S~ z4x9_RIEZ91qVXe}gf{vN_D7hs2Xz*8@59$Gex~emHr~ zb2fEm$4*-aB@8Z04+5kK!q~sqUN!IUQ98gTV~V9LTCZvx9nl0Lzd-FdzJ`dv+-{vi zEf?If*5!^EO|T?zbE7M>sfh0cNY3BMk>91)Uoz4B4xS!=ntL}(pbb2%n}=;B)?;XS ztUrspkOoDy&4KGz!8l?c+)$g);N8W(TOHH@`+q5r{ zgf(Fx+;pqt?v9uRb8x2y)p>(T6%LWZu$c!-2#h_B0=vWidIVps48 zum@q#?X&P;4eON|(CKo;Zxh{f+$4^o!s%Q<6i$F+G=J`6KHZW)FI6Z0j0Ud4&qA!P-k@VzpkCY3<87PhD?ID}sop4ap5odj z-FxOZDO47Td3>AH4ic+9yrhlz60mQXnEQo^+SVQ9MyW-$>E)Xi9A!<57MF+4^eMKI zDTI89wWN{XLK(EtO=c7PCH7;IYZ$xo5SII*(c%HIYH`y-wwjcjzFM)Yli{P5g)^kA zuh(#A0q_CCu~91TyYqgE`fuuhMkb!1v(Q@e zy*Ai7VThG*sey9(>nxB94E)&=DMJ6LP}zkX`XXAfh?`nx#k)Kg(1<;d%9#X& zYb}f-(0aQPWz@f29=TcOZNuZW0|n_M-&lMEtSj-AwS9Fp)}(hC4y{6Ox=|zajrB{x zhZXf5D)0eSWvPi6Evunji^l~zfpV=JF?_*}8L}YbC2l0cL+PCu9TvW+kSK6Enx4gj zCzqEb(S-}yGw%-WMlP%K;#*w=zBJ@QQK3bJK(_{OBv#0J`}EQn+G-O$cL&jmRt~&- zNVy1j&5w9~r7f;x9Sm+gx`l&lJ&6cJp&o}=Mwy|qx855O=XUm0lBH8|xq{}C(vi*Y z6d0B_$FT_Gpj0avElhPkX*l8fCpJwc3-N(6+I?+}kjd&aI4`7VlgvG9Xdqf=HNmQJ zO&jk476+l#ylrK0G9V~nxNC@HAXOP>;P(YyzE_$8uD*OY93OeHR}&;zGXq5~c)J)I z3mA@;rJOj+m&ATpXU{S?lx`EOhto6ugMvJ*NE#rZ?%4V_S*ON@Smgdg1w73^oC^s< zRBX3^nM=y_$uWOAwe(t{(rTtr@RaB+*3_p~e}4#-*V%yH{Aj6ZS7KS)DnV!?UQEW@ zVe@arqaJZ473@kA*QN)<8LNf*{AtVWE3Kv`Z(>AUl5_x*0`1B3ag!jx$01v zhsVW{1M_L#C?`0*Q6SNCl!jY{GfXmwG&5O-rt(y85yU(w<%n4K;*}IhKeh;hW1f{7 zIy~99`GJraVmBAgIK&V(I}&^%B{2`^vi)93xD+$-(Trl#BFJqCN)g|alkxKGABi$> z79H&-1+#H~DwkB46-JdD$GW$L*NM!NH4McOByg8fsRbf#UD?BJu)?jHV`;fJJL?gJ z4gt<24eZ}u0NLE#dptQ{&r~&NPjj6KWmfdKbLhR2uTlF|>a9b;Du=a_?Gtz>nu|9P z9L_xdns099^yT%Jhl3X(01}hm3b6EvfAW&pQMx}~xu^?#e*S(>vRXwz$mN?w;i#Ot ziQ<{H=b zCIdsQxcZlJWafOT)K}wy}{z^%Zc&k_avwJ!5-uZg|E8i1y{hU zZ2mt-*tPovHJuLmE zO)5OC(Ksv@5@SB8d2JXc(ZO=j!jM*~Xq_CTPG-e6zG`oan75^qme2%8#Mr`m$+BH) zwEMISKfn_YR&obavCa~swXlq6x|b6VZxm==5B!7;@$ar&O4k;sDh*lYZ1kLD+u$GG z{}7F9jC_LsdkD-<2e_e^rzF#<7;XQBy1~g%47e}43$Uh6^jxFv>0LDcoM?ePNN6$Z zJVQdqDv(C@V)?Zmyvd#R3x2Yyfh>O=-==0f*BO z<%=D1;go5uGFOyz0IO#=>iH0f;)@zbY~ajWdEms}lQdS;4NrQxg&0FF&dH#C@@OUa z?6`$_3NIcdhW29|_8og%*XlKAjsco4kE=aIQxxMGb6jyc?gR6x3`?Cse+}4D5NUf5 zN~OyTHV6ubxQMXIU04$)O!Bkx$nC8-hr!SCOQelIFJq1U@!rQ1Au3j!Arfh5GLw!h z^yG{c>YNHO_%D9l{T}d=yo^sbTB*Inm0DF(JRQ5>7F$Da_F@OzMgg)D^trao_|ODK zwikYZL@@CyvJQ0#qK7p+=s40(&9JBjo9vv!K!y|?`INF9PsLl$g?%HP$*r(!I}a2% z+y?qxx~9R0i&CBf-V5sVu2y?2WGQy4k+oh<{I^U6z7~JIe};6tMe$xPO_ns`cW-+4 z&{yi=e7LRzPL{m=WCJxcAzPu$OaA9uh?i}E-k(<$`+d^39bR=VZyGK^`kDvHZHQK9 zyjjsLtp|mz^@}0$aw3Qo*xk3Zz}cmhDG)n3e41<2HA;kxSwWub{-R;OCYtMq<*2{R15pl*5Oa> zObe%`^kMTrg`>q@+{oNQdbv7ulht;@xYlKtkX0}zWl6;|5y(B4tnK(uAhfUC9a5*k zayCFy`0m%_wP{J0e{5%9kC|%5d6Tw6j;*_c@kPLPy`*r*nKP7I%aCo12`}^Rf{3>Q zb;M&J4AX}eR|mx~lbOk<4R{G(l;eaD>%wgr~`wDS! zOlY~fwpkrlNO29k$MR!9B*|PT2dD=LYcH4G7cQKtZjfW~68VVh;dsq{D3eKBEYEN)KgQs%}yNbi`i-dte&MwL11XknR7$Mr5(|dHk z{Ln@AU8&4i=sO~9v-;iP$>RK6U>jg(3O5+zjS5pu1sT@%5}d0&PgQhSswvy693OBNVN%-G?HMltnBfD`fR^+&4Hnr{IO`&~y z3a9gz8Fu-$xZs_LX!*5096xA+2joMfGx#lE6=*&c`=7DIQub*bMA*>axsnfy4&4qD z55h#8z2dEb0NG-0&h&av9%s(hD;dX{B~Nf8!`$Py<%sd&5%(`uKM@z6(z}q2e)`3yetZrK+MxS)%vTLwIkNQ2-S| z!{hwy`G%o-mO8b$3Rq2^)W6*Hi(R|aa_y5kgC#xM;euzr5l6e3OuKakm)HsZ2^ZyA zzJ+Uh;I&?d6<_*%KEZM;YsrmhTe$)Sx)~~BIzp++YU{y}S0_p`;Va_(xIj=CRbq}gS|X7 z-W*x1f}g~JlXKK}ARnnmAy{1MhwA(;pA&ulWHkeDUy?Zmo_7Qb@yNv~(- zWQt)1EX%48JDo)j*s1o zdNf6%WPj^y5%kjZv0cm84oY5nAG{3YGPwQNpR@*f|_)4LFX*DrZz2DiFbf<9y(=RjBM1k*p)a|`t=+e&pJG1%cpMB zWc{tvf}FseNK6EXV@U)>S}BgD!g-U>sP03j*Yzt5;_2ii=5jSEbwnLosg4AtiZi6Z zIy-~?de)ni+u3W(5Akj`@JoOcz{;$J&CuZJzVbX!iJjKP>@A|gy>k<$!06J!V|V5% z^{!gV4J1haLyE9-|uh7l_ zA_V_r;Y76UQ>-8b`{dCIj82pJwWO-8?M43jw}QVV%?yzM1?$!&+&T;-HT26Fz8flVpnF;d2 zn;}TG({nZ{>$}a7ejLQwZ9nTn`}0M1E~hu>*NA3U4Ops$NU_8OMO;)6KXW&{%eF?S!t!r-5o z)+r53V~Jf4!xEGl14e8&@qUsM!?(ey65Yxz+VYt)b&tla{b%`SJ3Q2~f+e}rwj9v+ znD4o}g5zuuy{4VuTa)4%?lWtpkepf4+?7Y+Jlo$LcIQ)hSe&8}pelFs*CVDk(TIbfqBT&oE^G z-%KSoSVXpsE_PJ?28s4%yNSrr!h0OtpEl?mWkZcbV*M|mEpEHgO|R~vb^5cdZj(bW8uLHz35WY@=#x9N9mwYxdZu$NdWXd1luLuw3SA;5 z{b_JHx*3Q*rYk}i{23Ky5;@q-(^Y5B{Va+_mULmm!enVTz&5JxDV+l5=2M{jP!nVG z^COej zY3%$kCIsOXa0za>?7rid63ehHPx)cB?y}?8OS?$c^6^wTavYJp7R$IxGUR#t(?o#smo0dg?(>;m`UNizH^CONhHdH(*qMuC*1PkNMV+42$xXU z;|2#%>5A1Isx|rT+d%v7YgqfaS8Y#OfXgbXEf)nUdufU;a9%wN+E8Tfynn&Ysq7S@ z^LsI@^Sf(|?gVNm)Vw4RBRDuHkIv9hv{ogbL=97KWF6`&3fKhYx-Vs?|L@D-oD`G z+}S^YALS1*`}a+;AUlQO4U}5RDXJ^tYF;}+?_*e(Z8gy?Cm5lOJX^qX+i1rGeQ%ClP zqc7R$+6kHUg~9{mb;n%^o#Ch6CH^6FrzD5tvV@f)6!D24=6Fu^%VWu=;ByWRxbO&I zH{3o_!mDFf?FtZ;NgW^zjl#Gm-F?xa+^NN_@Or-j-iC2Q!U35A?n(`bH!@{!%V%jkT1c3ct4_w7x$cMq|RYK z`>S)a)i2(HUE`RPgUjTKa?_>vz9ptVjjpNjh2&7vz2{ z`nfI-e$9jA=wSVR8?5pq184_;w|c~4bI5T>>vj1XvUWrFTaD!%-ZB+3{^?)<%bHxh zT2CJAtf}_U+4>|&=x!OxQX!mu5PWqs^GEp#>0;&Hg>T~IKv_S2n-)BQKhNM3S+QpQ=%da zL6m&w*Qs28Ug$^peILB6LpwEsi4XUs^&8asOe|d58VA3*iLkuCgbCfJ)kJ7{u3FHD zH+96h8TM=Ygxw-v4Ci_I8d8z*)2I&rNl>FSziHwHc(|foHlcWw$<5*l(2?CSe_Let zFBVTe^j7V4k`h;T)qZ&Jv0cyv$0y*Ban;w^IoR%@Z)ex!TR&Z4Uu48IUgm}Q9u4bt zTVM@^II*nb;%l*r>7U9#@Ke(mFPwj92e}nZ)^FBmKQm;$IxjO{B22y(H~!7{%q;&{ zzONy36xl7zsmf4i?X)2lMa@)?exfWY1TI+?|Cyi;|`g0JIV&`BFmDN^}s>tMsSqZ2;7 zn1-^gx}f$E+Ddtp|>#21}lCn45#! zWg;bds8?&KKootD>n@J`TFJB1*Mjs;WX=^|v_3_#9F+|I#$>I35&wQ>Y{_!5#&eC&_x*E$1 zJ7-kt3FahwE>yF+RP@I=AC2K2PjVXpA&g5hJ2+bgvmC^ zT!pf@6sN0ccL#jdwbq)9##qyb1vzW#+{1pAlr6CR2uCYO7%l6a%Ld{uwvK-X)t=|~@OwCoV` z$frcl+y|b@l}Xu7<^aw>({|TT(V_G7h;q75MIOn0<>m&z=vg^u`_5i5`EwuUBqL3| zPh@JJ+TV3g$Cli)+)e_0H@&~_SW`*%)a`<-4xB3)A;a5G9Xq?o*JMt&&!RmYqnB;& zQ}2{f2cYDcRW0+kNXqi7$baJLXDJMOhamdJ+?OUE@yHE2dfZw#{yR5034h zHGUiq2KH_rhX#}jy2*lM;oO9NvmhF$0*S&=s6+fNKN#JETl*7R^vcH1NIk2>$Y8==|5!|EN*2IV z26#lDwNP@jq(EvR{?iHiqhIF9T-qdA!a}lu6SkjKnXfxR$owELcs8C8kzXs1_M>Bm ziGR7|Rr(F7b@4pGWsYHNa20b+Pl6WZB-Dqy{rcP{CtdyyO5FK3~_ z;z{-w%Qe>A7MIGT!h?9uEd}c^R?07F63u-9Tz*=xeQ)aAyJ(-!WP8%$Mkm`x z5LEsdqN8{(?-ZNG5Ca#toR4mN-}yD>W}CY^?toZs|9iOy-r|KfG5*h=85y_|SJ+;N z(%K5O+0jb|By*-Vc39A#+BqgB*-MYrRq-2#hiItqxTrRh#Gf>!vG0}Sis&O~8M$%y zwR`v!Cu76DQ}N?pVq&#nL$YpxGgn94k0rlqHcQL0BMaQB^`L2=-m-)rtjyN>@SeoA zLL!opO5*C8j38lu>dQx~nFct0ggagXy+E=CY^lQe7$d85=~|K4NMAXSQEMabfRh6> zuC&ZiiPknE3M4VfOuntVeB8wQ zBz`4r#D?WZiaLjQ-@?2e`wJ;u`W`wcx!M|0AmCm+eD!xa)QdGa73j3LhOkp>)0 zy+$|2?Pwg3yWif@^yGf`j}ufK^b`)yXWPiL+7IZE#lvgRYB7vFQ}Fy(QejaBZuoF%w2Zz@`;BnG z=^`Yq?i=%*goH*)y(n?p*eRK$?8n$_$#&+bS5XENR(b+Uc* z*~oHZ>R#bZ00L8fOBnSSvvznN*`(wuTB4~5ZqHUp4v%mkmq#`rvlk)OmS#97ACGY< zih}0ULDOV_V6 zu|7*T4Q1V>&-CsZ)^nG?`U<0Ra)1*r1?^+Um#wNjQ^C%LpCQrtJNmR8>m@JTq3||s zce3bcSLfc!Dx3F*_N9y{$m;*h3K(!eGNJJS0XmRVPHNw*7s~@=&i>r$D26R z#>dWkN8#}9n_9MC=5r>8{qOpRS4Obf;2*5%?}#yxEgYRPzjib55nSkV>7u@)qYjuE zB1YwXAM3`Deen`b6l^4z7$+zCpcAOY+^+f|p2=Q*Bp+z^DyFd z`YFTlsyr~a{l(0t;dj4VWc?Hwa0e7ITZD@MhTRrLmT8W+W|-BYs}#Iu7!DhN!>&J>PxmS zq|S%X??7|^t`XOt(3NI$thbfE1-roMhNLAXTcOA!t=Qq0azhvAH(o+ylrvBk`w~gX z8Nv_E>^g6}RGn`S#5+MZuSC_-!ms;Vs#0!l_m>b9w#A3_k>KPw^0YM{MI!EE?W|^T z;PZGMqGbXf82#M>*t}olc^b%(dO035nhcIDojl7`>k#Y0T7q)&dz< z=!v(Z&EjB|MXmZuJks5dPJKqHbc=LPk2rHtR6e&(Oc4J>LGlsV51>xh7XPO*0_;TjM6^nqK*Y!9Xslnag8KbX zyKDHf`l9pr=mDk&^X457Ce8a4&;oDdLH0|5;9)b(HKpOhb`KIO0je5AYd0jd2js)?%_{S@IwC}DMWcj{ zS@S!GO#%s?T%AChaLu$J{=+|&%?JEh|teWbufwdONRF!DIij0GrSw5fk+h1pczRr3%LDw(B zjIOQ84EBW$QkMj=7c5)&%8i4CM+Gc|gvKCiVavq@Tj8-f#Y!ZTlkZt_dx}0DYULwj zKG*^-TCoi-XtQ=bIG5m+6PFExXQoCDl9daB?j`O70EjHEE^JqPHx`5F=UaxU64o)h z1C%mS5~6=Ot=^rYnq+{Kx-tBAa3)Qh8lPgDSv>grdx<2Yf`C#LZ;eCXTbCq+QiC%H_=1@M*7fxqOt-rv!JgVVd|XwH&v% z4y4B>(El($`Z^ZnY2Bo=Z#SOwwz}P3uRn|XS2}XU26`;)EGWe^!YQ!9u9auG*R5}# zx&=E$?+EyImgShZSReLnfFZ0cuLl35$lEAdfgRfa`YohD8n@?=&N^@tYAdCa0m|r4 zJ8u?{5hOhH2nKxbyLCu)FNJ|%bZJ3TdU~9of)&%LTIwtSBT}f}|NEfjLqZqx)~!S@ z&q4!sxdbVmB`7Pr1E}}O?WPDlXvw#XFWMg&b*G9oY-2WPy!!dFf!oW_>{6h>ZJvkP zAmyEFO82KOWDCOj4KB5w1d>-X%t*n|p+j zC{|Au9s74rgh9%?xX{>7C_05Yv<;*6k0UxemInlVd-=sEt4ETzSWSi7-oE9getgN^ za~L1^Q@XtT0qB&aKrO?o@2u|>0eZyiIY22xSP#u;yMnuUzH||7PN^}>6~J7?6mtUz zcoCQi57(^Nd#sL%B@+RHdYbP8%o~L9?t!q8nPbk=?53H|_Uk;yn9dnCrjeaf_LO;; zF7-Xo)4Ou&I|USiZ{+PnsDa7>CYUqR--($z_~RPPW~Z0u0P9VPh@S1u((Ry-Y8}V8 zDhfR4HF4X4k7HFsD-fY|y?Y~8)&=j%t7%JtUb~FeFA9J!X9Bq5Ti*?X8kjoBU=+cA znGLH|w;XbOFm8M)j@sbV9!=JpZ>*XVFo3gC{ov>}zB5_fZXk$8c%8}}y~kzo&ZQ_V zsTqG2V021dnU^^*_^j>5!)Avn4FZaFDwl&>!)>R2xp(#Hz~2oLX)q)DJ>4%7rvI1^ zbkR(KqUK>FoyADpPO*Cbq0ng?b;p0Nc{y>sIC#1sF{LD&DdW9)TPgF%#V z5XDnQ`M)31gedsmuEZk=7cemb0>6aaJ-B3d#xLI&VkF*$)hG2kxjj2@XySMR z8DanxfaC$+lW2iTu=%4bVIQ=s6yqsl*N^L+pfqYYrjix+Ih8dEL*@e@qX zUfiMBb_lB+Xdf{kw2@{$R`GolKRvzKOr8xAa=nVn(pui~2uou*9SV1_Kh4`6O*dRY z{$fbCUqgeZNM}nW>gYHsrJV|1rlI-Z;X`3d6Dz3n^z#0Ba?R+UPA)aTH!MS{%0sjQ zhdG+&ja~l0*UXfP4njc-Ttld>o9HD&!`O5@ZGAuEc3AtX2uYxW- z4*6p+QK=MtK&j1w>ZOvK1DO5nd|GwS1bK<8>Dvo8N0M2(2K>y34ao`?A@<7FiU2IL zC=7oRO^jNK1-@F5!LiiNx(o`=7wzmV}CmjhPa9Zj_%UXjJa_&9l~#@XqUZDolo# z7_eSn@%oX~wwI(fCoHZTR=DFYJeLFsFrJIJ!uB+B;GsP^vhz2m`R2}SSww_7nTX+Y z!_R?YiOeA%<{Kf--o@W%`Nv)_RBqhMM<0!f@5O!oBf)ZUpdN6QY+N30_;!~}jgA{- z_6W5JfZAqW+2l}-7eRx?#B~FVc7x_YczdGT06of#s@pozePc^u+tr-UBzP}1IKQF2 zO!POarr@KNpCL}zem>^DYaSTT&~b%^Og?;f4~;fqld%K5meF6M5Sl*q0GPai-L2aN zWW+~Z7$tY|^hXv(_KT*!k8FtJunT-e*Yy!yS@Hb(+UU9&Al8(W3zMK$(*HE1Vo;0A zIHPCQkLYM`{yB1~b$lw3*-yDf$z!-zZyF{SuyeDxWK+5Q2Ays68;$%CD%2HGp>iMD zJm#WtuN%YsF^(zYXl_G6L!#Li_4g7c18#h1aOvU3uRa!u>SL^&GUXXxd-h(7p!1Jc z@bRJdg`U*z?NHXq?B;>Z2t@|)fk%0QYGl$wQ;)%Zt#yy6rXcn z3A&+JqVeybh<@l1)r|IL=yLl1=z7bjIJYHQbR&Vrg1a{s+}+&*B)9|%5ZtwKf;$bt zEjR=x5ZpaD1Pj`@H{N(&=iKqexbK|(e)Zq4$Evkz)||7d++PnZX@?Db)*pp&r3v&U z7->c2YCsO7sq(dF6@yI(527z8pwD9pJd2h#{1N&^N!qfBuP|lxYM`#E+uiwX(K!Z_ z1(#}f;Ner^FDRsD_J~OcCjqlpn}F$Gf3VL&5r70ii}dC*BLgn>*Alm)Q978Q5x)bB zjnetyIw0d`CBx|JsZxu`z`1XS>8jy)JI2#KqFy1O#^yz0(tucDjiwpto~af5=mC1_ z*SWS${qJfp53`nK@U+g()=k8`QvtH84`@Lf*GJn-{s~v0JVGl_D2m%bB(<2V)>g`^ z=tF})zpQ|X9}Ys;_Ru4m+FsPY6CX)!A3sECBXhITT}qs_R&Tg1On(&6Jl~u-hRY=!Y1XZ}WMlg@BSq zYQqw2SzH~y1?s;JTFQ+s$z`An>@I)y(1Y=WB}A5fwDoO!#4osdO?!=#NRND;xD#ySr((VY z^tA^X?9%+k&}pe`)v^df=%t>rU`9=d`Zeh8KH=odI@%?=E?T%`VXFX=LX1cz7M9 zHqZJuzi(bD?xr{-dlv|OvRNZ;hSbY`))s;d>p2s&KCV{9Y1&xWBNd&tbFF9X5I9Kb zzmzj8dd|qvkav7d!aPW$3$5P2+96yJzen^5?WGfSIJ{IQ10_Lr4sO_r+%^<}oD*&f zeORndUC77VD|Zv?w+UAwes=8Z%ijyRLGG}#1`95UciB5{bX4q|dG+TVtu!GXqq^YV zQ@)BnlWZ;+e||VG9J!uf?^@y~ms*cSnF&GH?Y&fs`oywXINmC>&&XYZKjo7G4Z zTDOlqXb(N?$o6c$tnkwV3m8KoLvb?KjsBqmj_0L50_&~@iLJDls-BkVq082)@gR=0 zPkSpL$xKThrX!>v`V1=0NHO2CZ_(io z*I~t}XT(vfNPSXOdHH7&jR6{#<1i<78QxsR%jMkO_^S0vO$}ai<~lOchn6B`0Fy!O z-aoad!HK`4Kj<+vN`MOVi2!MDfzAXdYsJ4#o zm}k7cG^pP3bdK_GSQ7ij#7PKhEctuMWrL?K@u{yD6!B(a6gwYW_&U7vmgfBii`(Tn zs(0rji!@9yxSMjD(Y)}({#wicXVd*ELJ$(*65TwEId|uY z>Tq4YHK7&%KPc@DVswO5A8tP2-7b^dLN8sIk-`R>n;%lM?i?fXryA)u82e<~L4yLf z@_;m%KMX~_v+f-uUv2kTuqDe1R-!Pyg-x(z8O5_Qc1Qq$2%#O!rk;PFfmT3O_o=}0 zmB0u2rQ)SrZ8_5CP}v z((BIHCNjFqi>ZRv5w)p++ zM}1Ux{y*i`$wsG7hk{wsK zM|Anq0sNXe%U;)6gNVZ-2+uZh&?mx)XH+-oMTMAaw_F{5-SJ+9y4#3Ye~YJ3E)- z=~aEz!8WqtZN~8ndd&CS@?+`HU%^m>%sTLD*#!$K=|Eu1HPEftL8sP=okxG;2f@}v z-#dSiuC)-we!!NPmw5-3WOE}D_*nJ(ba1CnYjbAP3r+sT=q3IsE&X4ar1% zsDp@4+jkScqm@bEzTOY}K1rtf^IPxEMasbd5%F3fB*y-tC31&dvvzf4C|@8hDv8 zpuuIrk8}8QhV$FRbZp!&$F;#sH6fEM&|#;M`gsLSity6<3xV;L zjW&s}X3Uyvq+)H%fP66R4vESneQHdPJ-UyB6?&GNMdcN5M7>T#ECX17Fs`Rj2Pha% zL<f zzZ=7ZBdg%yb8~U)XxKj+7~d?m0sDncpsP=-sN^k827+sQj$rQ~oTrb5PAk24GG99jwqPHRWyKOVj?CwB`o@-efyFUIDGj$`yVYtUVPaH^Fs zB=gN^y9x+PStacl(svf%o3K@9WpALT#7^(^qs(7rlFABtQvLHoli|swAX5G|~%lscwLYV?cm{rO)`xX*8ytcFkg3 z=^>LD*vaT?e@BU78~(8_X~?*8weyHc5v4yDltVoC{E$rgkh>(VgV$dE6c*tbu~?zI zS{rr*zsof5afBMYs@bYF2j_Cz1=xnPVNdc+K|g4IQ(9r>`^5jO5{33IqM z&B{Z(j-YuDB0J$NSX3sPO+cX~aX#kMWC^M>EX;?kdXOwbLIya9`?+2iFn1s-Mm{8@ zNE&p3ef6$!@Oq9MAwq9*ajv>)wxXe60QwyDhApE0*@I;5dtuVsc8~VmSI&EO;ChT% zYOCB7=}82HmJjB*zJcaK6F=qWx;b+27`O1n*?xpV#PBk&V@Fcyp2e4zj^l>B*MI$K zpX%MK&OTw!oO<*}XgC$z6T`z*RvJgY>@HHYlCJr1_bvx}PfM>V>BD@bbvUhpScFC( zQR|r+ba&>?$=80+UoiJ3s)rvwG3jX)DR1m!hZirU+zo~TvZ0#j6g59j|=oy=1=zTdcPF4edF4zSw0@eMvsNa*vApyVHy&b zoS6q(*jhGzHRS@|2^(Ig!=NTyQN7JsyQg(-Qa4788ey0DD76<>_uv(;lmc1rlK;3$ z4=OqPzQ<)8LKKM7T5>xY@`Z{pgb(@5*VJ;SqtRoNEgAZ8^S7liM?`<7$lYlRgD6|q zhR3+CIgoVLd6oy9z2TN9WN|ok_D}0UW88-7vl#j05zx&fjN)>S=Kb`ssT927sl1dh zf#oX(MbSlW5GYM)I1!QGa5x1c${|9q>^cfL8}?ayVG&dro(>+I*tl*+*w;DV69>DgGZS`m@@LI+uSsSouL)Luse-nlOp$%#7ag}Vg@FDDT zik6ije41n!vDg0Z5JcbW6@NK43*4cL$nd<1GF}=1kFXZw;Py z?*NFubLpMY_+jqaLC`h!moG`80636yN&S#GvIG;g8~`hB34hFeqko0i{<_co)mbO1 zG!RXR9Vdpd|1N+>?A$7CdT#fXesCUIVVsf2Vl_D~tJ-+X_P$c+7iE<_{QJ7jk^5A& z){B{$c9Q;(76#3Js{UWotQq;!=UuNhoC)Q7;-!wcfly8ldu|Q>Fr&*f6{q;awJ18@ zDyx&e4@2bK>!^`$i8H$q3JhKIg)Y5Me9wDNfr1P19f6W~y?F>g`Aa!n&VKLsENf#w zG=N5aa52AgrxLH2-l)Wx_l+Q(<+gOq%}3Lp=}X)`@KFHb3MzfB7^_3WG3^^#vu zkdpt}th%yCc3p&NmBH=4!lcs<40RH`Q_72JT~304rx#lO7NUwt6GLEZ$h| zy~LHWdQ3#d*(v7ya8khGsBA{0SnPe2@}Bq`8IAB!4!@x|@Qve&NCdbnFrec88TR*B z^YRdDm+^Dy5*e9mr_$5PnD7QdN7G?ZA218eK8*~RR&g1f{h+q_ym86zTE9G)wckhriBVVQMyMqT!jp=aeGaeXlVD|R^-vZ;-o7dFsoI(j`2ElM?4^vP_xyB z%5MUk+b+vhpSEla#Afb4Z!HP+VG}MZv}$=*^T9G-Pq+Q3D^boQ%Zf?|h9QSyY&$uI zZs?0&AwKP}X|eR9Qe9*6y&E06tknhRcDr<(97(xYlAP7PUx3ndpEuMzJMqkcYtFNyXV?3N3CRG_ve*`{T z+HZ6va1eJvP%6Zna%9tEqQ6H*`D4!(8p+(BwbY^&AI>z&k!+mGogc~6N?%j$q>VHF zc?SKuTo*@#=}!u)$xL0@`&wAY2xvJucn(ydFFbjLoRg}d-8{nX@xYaeAfxpqVskMJ z(9JhxOk^SNmIJHr-jL+`>blL;7)AxtCSv#-`rp!lT#S?^PQH*(+mQrDF&FWdD+~n_NCJV6`wv_JjzwGXxbnNEen@H)f&d;4TUYsnUm6E-i zCS4-MvOtJ`iN;Y6jh)`8QaZNSJeen#Ka&tlnzi^R#C`+ko}m5wnI5kJwm63CwTFc3 z3v^Z0@Yg{36ihKolvgq253RacGm>|z+5Z)u;(JLBUN^lny>Rsd+uJ|x+p8}q2BXyI zTF6O+hRkZ*hPW8eq?|6KM?+z+3_9HeY2AXGlFrP$~U&!pf z7-}XmZa^1*$v3Ge>jc7pdeMAsW%vao@^V-G51FnR+aRvpt4oKiq!;tTFH|2xn?7`RD9wFa;^Ugy5Y;ZX|h6+=y-e-x-V12&q||e89=DOM~lk_NS;y zbI9$nALR7Spm(`TLFsa=NgPMzD>?|J{_5n}gu8u9zhj>!I6LR~%h|^SK5-K9Uotz$ zkplKiuaV01KK1*}eYx%w^As`N$z+wgGC504T!S50TXFf#j1pWADc^zl~t7V*^w zLp2t}0$M2<%xG>W+xL`~Djj!9_6 zs~gjYu`KjV$6(gm5Kp2}**PtS1x?;l?FR&2!ji@>H>@wO&l76!&hW>4Y59C!JDU{R9qgXA9u|bpzK#pY3(VHR7cN0wKRi-6 zGf%kTj$60oDAfE)NhQmlhFZgon-8pDdY8Z9x{d;N{}7&s_U8Yc1z`Ej2Cqsfqw|i? z`uG~J#V=TEKmC%fyUxEQ9)h+$?oZaZ zsvFl4jK)k&5r9BHMqc`RRSef>@;kJ5IFb(W**$pjbxP|7F=0wm8W+i-LFP!NQ=8R< zKjU#eJIZlPA#W|AVnD*tH!|aC@l5U02=1TMacfetiZ}Ft0T&gNhE716KI$pFn}2v!4Ze~v%zzU|)lElyPR zR`;i<&wv}QxOIR{Q6X_vtX-I{t}47Tl-dH;InUA?1PJTr35>okjr+CUv0OT%B~Rml zOp;c-Hm7cpg2y-0T_>9m_iH+nFA}+s{LY(o zv?u!S{eb%$)~tug{3y&n3r2LEwLc{0DmCOLevA8Bb5nD^&O$vX*HZ6IOmjt;ab zkSs#Y|LG$b_`Zw!u8FuNFE3Rkq|qrE!RPd7zV8dZIHfp*%j00e8MXE*jwh{|En+OL z3>hQ#@JumdMuN4YMxEyUc*(dkgyhE3_Gf!WmU_&Pyn3N*?iv|Lby)a3E+3lY0nDjwcdzu00 zY~w63QpkXvfyKjGi*<}hW4hUrXubTStZaqYIHMi;0xw(eD{&c^6Q6zc$UUdeF`7JY zNs-dkQNfqDtXFsGdpLXrBpI`I-1%$VMCPh{*;nN{wHhx=mQ57C&j)wr4@FrUI>plW z*JBLsXk-ns62QyH{Pp(~p{QT4G5gjlw<-(t43dFehE#^_C6}?QGt&>Tugq_6S^@13 zAwi}nrjEGPPpU%h2j;aW+v=2BGc8QN?VoX`V#yU(_!mHpzagJ8p0Gj38xLVhhlnPf zMk%S$8h=FHoj8R4g&xiI;GX%&QQ zkR;cxkxtXO=87qeNKHwGnenB~gKKG)2~OLkicoH2TL|1Ln&SIWTo_ZT`9Y z!4kj(RufZOCgp$4!;|Y*+d0L1Mr^A&UxOCtbV)|`vvVdw`|8LVJ_$*Uo1rxPbNgH0ICIY^a;K z1;L4mVK&!~KOTV2pFgA>D1_R?rF!~zOYaOq%^U_luHxoHvpcKp?Tj!iQe%HN`PRUd z>CuFq2tR1*TCGU4J~4FZ{m(=a@QLi)i!~x5{miYx5<)XqYR^uotL}>gnLO0qM9}&<%6>94dB`-j$c;2gY(`;~%a4}Ad6WHi z-4QWh!NLbX@y~ssXqRbiuN2waqf}WNIT8ig=W66i)|sofv?EK_cq+`OAsmF3) z$^7s3pYy3-jOe|+oil?Xf`0G3cf%E+`Nh0lrJ+k@93I)4lDZi8S}Rcz#m*tcaq zD%%|Y4^G6YcP3nJ>5q0=vaC1j8v*FD_XP*IKhMlRVb-enYt|+ z>81xn%Kz?JyB{!>p*|EQ#5F{=Y`=nh{O~v1H&7wQq7!|wc!Y_f=sGQNG0&95%9f$Y z(}*LJB(T7jN{k%`o<++arOZ{*-dtwCCeCaf5Ctrg(q-pFbooca&evbMo%J`f=9oEx?DB&wiO;oHV|> zx%2!yoT;N0y%{ap$B4t!N&S~XcAzI}iK$}Xx&G}ViQaP!$6sQ^UBxp;Ode0CamPrE zH*0g*oVW&?g7V{{<;kkYp8)8>bQFMdst*dIbAKG&%p6~JS8a#Ja$4S4nOQFF6gODy zUaMt0Wc`w;4iSd$8lUOk{8%-2Jw&m$sxni_o5$}eyG>Xdo77^Lyecebn9F>55Do-~ zWv{p^?XV|pw39_0l-`!F%$OF-1;QcZK`7{;zWL7F2z)GM0^y~_vy~*uCfzSH2H{7! z(f1*z$kIIQ_22CNE86eteguM1}0nc-Qe>pcTWrrM#oTT$i!ExzYDx`J5te8s8rU-04A>vd`E`d`}W7O53-5xHSIARQw1Q zjB{d=flsg;_WHH!FcQKc|2!d=4^N51+d*P~1Sce3jYx(RRRnJNK9PMHewUE{{kIZ5 zwS;V7SIEVyN~8(O^limk!V0O4%&)NM&O5fxXx+pJ6erueAU=j)e#xB^W-undR#pxC z!&pC^wFWOohe@?!+U_)y7Tsbki|!BK*9m^xt>6Po&@;vq3!gu=j>y*v)mCZCSvpde zanLd2UmRBY%{awh+cO*n?;xc-g|tfXke7JBrsl{N@%F-G51K3040!N}|?&Y>M55+#nXR6`^$tn`pW3RO6XdhzkWo$U{%uzq`$CeCA z+WFNR5n+iPG8wzJXfM(RaG4yn-niD;^Ys>s2DwYX4WIv``N1KH7jkkQJY0hdP$qk} zXuO;5MmzTnAj2(OUt(k&4vu`*sK#Pv;J!*1J+qmPG=zwZ@xeaDCcK5uB5T@Z=(->D z{q_}c1_E)GaV&>B9{t?Ohz8H>>D{oqjQEuPBhoL(Z{|z>`4SGQn>|TB(kfGV!PxYI z7efGC1c^wE02K3(1Ygdjn=eY-V(D$EN_5lSh04Ei~@b^)`ctLjwC^L6T`DY+WBr*eb5ern0oihnzm@NbdPKkZ|C_Zq$?3R8AH5h9&=K7VIYZ7$2CgVmMwEa6)?@%+&rI->)a5!faCHvK#H} z=vWVwvi3Xp7HA`(_dSwN7iHc93rIG)cR3!I|C0feRsoi1TlkWQXXRZ`S+HQ8`>e({ zGg?badvK`^obHg&!C7jXLboDj`w)&6OTm>jb8v2^g&Xll>pk4u@k?s+yus>ntimK> zFf$T+3{BXaU++D#jNGOjNs46T`(^fR>8;kjM?!aE)loiK7*ANOk<3LEk6X=%G1os! zCI19ASo{wAao%moDuYcHxAWJk6th={LA;H{DFPCvM;6IT{`~l0AqpZDUJqFf6T)~&&bJQDtlF_S|UgvJ}v7tF2PaEW^ zP5$t_l@oa|X{9Wu29!?pVuhiei-NW_WI)rm3S$JdId)ZFFzM%#>q;Ov#Wi0*`OVbp zIZwu1X3S%9ePm)$xfCWv!#G z?^Z7{Nib7P*sC6Ug$auUg8&OApZ1!1pD&D8nYL*)Or1d0V{!38sEJWKL2vHf)QAQVcLc#Nsmnl0sbCcifm=dmy*HhjS^23rn}|ZP<^C-= zv~~+&k_}E1Rj?oLR3`93P#ITBmmJjQT!poH?z*~cmBnwVI4AX!% zfcNIN4BYZ3WHH^Bz44-p{iSzhlerHE)?nSD&WHh1DxwOVpT@r>C%F8aWYmRdom%_b zX?}aCHe#({O7?Nl2ja2P#ll(ggTsP?Ulsrc>+>)ob#X&o+>y7)BtZY_*qwtj>@qAJ zX#zLDWcPHxMuxuE5NI*}nsB{peyHD32iD=DDNi>5E%(dHQK*t{tS8mkpKCVH{%l)w zYFn`~Tye6?M{1zF)SU+gS|xprf~w(GrUdDfS36v*26U>GG4ZL-DLE!vVtc|=4m^MCP$f7*|&n8Dqr68BV{A+4aUAt=ZbbGB?E1c6BYzbW1WG(bZOP zGwv`1nme2pY2xD7A` z{_}yPl`6LUaP%O%dhj>CnV8mS7GSzuYKn|r%+)%3#U*pS{A#BM@ktmMS~F0nO@W1d zc*FYp-ot$|x0F7-sV4H(5XKgCOZ>OR>;oPW0@;uPZO-UcwM#|*O51*(5i|D#kvgZ> ze^Q%;)hZ*V>+m~ocm%uwoZrhL#v9$`7%Zq3BgWOUd_q3L91yq6=5-#p0aB6+d*1%4 zYNE;Pv;-kH9Scgl-tseVXMA>taX6Z|x*_~| z!lHXtZ(61>F|L5J7pocWyid+6D8Tezm>XKT1wW&7=zG_N48!OHQnj@g>grTds% zAXa3|$@83%2Gh8PiIR2TR;S8e6Xbu>+4M|^2FZ5=iJ!R-sPlqf9XQL^m2R@`YY(VC z@dK^(&e;n|mBaqQ8ryKL^U~%%QKi>JDll^z{KNCTPeM%z2TmFaozNA)!ab^a5J&z zFB`^rvj^+c-*J7axT7-zQmJ)n?0mM{vp=ByDxoBC3zEwIZuZNWJ(&)1OXDREh?;wA z#Eel}ko}Og;%#rc`tuV_=Y{VAz8beVVR?R7`0)|4W;EzS7t;DvS`>v4=J*9B?OCJ? zy>xa^&^fldKh@#Pu6Jk%(*j))cY_qiaV?D!*(+D|S_=9V{+>tV4fadgT=Jk+sjU6) zXL2uA-(~P!`(NdUi=Fig9nSZ6qJ>&e!U?c1kmGIErkV6xYok-Q*A#c#-W)D3*h=hT z-Uq9;jb>BWInKW^oL{%&;FTrcND!9q>wkR-I{x%PGIarib??8uN-WMQ8m^jxR#VF< zAzOCeul*r^C>q)K{4o!|^T#3JljupM@*U~~(e4o4$xZ|86>>dj(gRl7N3X73EXxJM`VYK>jqU28e&>OAK zn7GI5+F%y7UZ@FGTP5`+#xqK-%E|ao={Q`@$6l zi5vEqa(riEDFJwYsN4>dQwVbANFBHQ{C_xTom`}T>~C-@Z8E#5M45?Jj1&prr1#O34nF!+ z(iudo7q53tNO}Ml7bb_kkI^4L9t>gOb2w3=wa{<_C#0-0bo#zLtwl;0Y!T)3#Be3tFxJ%cB}U%w z>9p^&K7HRbx_7<>X(F&y!k`vJF8agSZ+O)&!t~=(bRpMSc`*tiAK#okptGd9*Y!{U zoX6FdYsm6+F}L!2UfM@G$RdQpPb8x6+$es6(7v|YB=FCjh2k!g^0IKRP8z49cF z+PA3jgB`fHa5oq?N5QAPs`s>rCkknXfb1@El55fDz5%|OE z5;>KdL9W0uA7f-W5NoGDz>u6pV#sa6LkZkP?g^&N4>KeodRa#*V-_WK( zHv0P7_ii^LV*Cp>vB6HES->3vEb9ADR94()v1AmEpoxd)FwMv0NI7*@N|n)H%>7(C zn0i*}=1bax8=rhH#|gkD_!A|`nRktUDr+t08O$!0>4!&|R8QHB9LU6Vka-gRQSSSG zvC1@)ODm+TU!uq4tIT5D1MK|Nk$SMm1|9NOOlV4ydJ6>AaRqgUTpL~E zx@(xUBIF@X@$OeFLO`Ll%US2XY;B*m@a~2*db3aiX5&p|- zH$WW1rr{6F;oH$WVCo}=U5B|YNOsaah@Q{c>s`qz2FH%^*}0Z&2#(1UIsEWyh%^b+ zNsj$ED0R@n}AcPg9zm=InFyf--K-f0pLEpbx9g%Y5=p?v)>AR#SsIEWf z4FF8ROV=?99^6iP0g$vMB<)(Ukp!6_0!2$%)@+{RDKAe&6?Jk!G6B}p7oz=?DCvk; zVC!=3CAjTP>dg6Fvd~no^9&mDb8UOksX@%E?WQ8H=ApRheNRIFzPBL^w|&w>9r_Dy z!FN}RV&E1qu-eDq94X~6rdXTWD2~?7=kEV|6Ov*8=bfS&*jFRudPJGa%Q@+!hI3EZ z&ouWz56@udHzRTNmY=d2x90TOXd6X%w>ytsS_v6{8*4#UX~ZD`G{2#wXI^~O|J9WF z*V0``n}jz|_RoQH|EB;WoX?71DbTpCUt~F%GF#tK9cX-`ZY6g_>zI6N5Zz4(+^DJV zvy%I)8i*rsaIyKoE5}h`dkTmv&a3jkl8v;)?)C9Q;v!!ZV2%GUEW=6a9GoU8mM9seD^@|ucmm1iG;l`Y9A-{x?zM0ITDbcJ&5n}lQ5j`PC!-+#QR z5DZ(=$|o{-T^M2X@|jyKdD{NWx54+frH{d)V(~}Zs}UsgSET69TWh323)vluU=Nu~ z`=1&*_-=CKA7AiwHw{uyB?lFYk5&f}iqLDfq8Lwg^F;`4gTttOPb^2O@Ta4nm3ZSQl^vij1a`GRRy|<8mCwf(u!t3~l8#JoLZ5y1 zWN?%JB6nw(y!zh$Jh>usjEe`8X*+HFI_Ep)T*MHkknHaABW$MxnAq6Z1@j81no^v= zZ@qe+Hfkk+zQs%}+L8;V6}-CjILde+6^U?vs{v11-(6InF9B<>Lp>XQ4n3Zw7;|>+ z90b<4L(|Y}yN;%9WpI_`kq>s`*VkA3nwZzEg|StTNn`I9C!DkW@Guff_R<<6Zs!Jp zi(L?kv7OgZ0xmIH#7;k*1G32}yL!A2`!E-%oq`%W{s-5eaX2MHiAkkq6<=dGgFD{!4-kmKO4xAP39DvKdITFccx|8?O z8Z(({WUo!u8w|Iz7h4fNWF7u*)CHYmW`!)h1Yp0sJQIm?BZnfQiz>a)R<5+sh&?+L2SAa8UUI=&dnT7mJ(8*xM zs2H!~V@8;J=R`|$Dr?xj4gr;ZV7|FfSd2ec7%zRSahe2$A<6p$YiK?ZZq#r<LJw8Ugf$l`pc?` zZMH^`Ntln-s4)pit4u`1=(csg*s|H_buJ@8h_-An?R|`-9n}rI&tbJV9q?CMyvl=3 z8Z8TUwVzWeFnA!pg*Yj{1unB|-Uk-#GF+c;tdm8U&|OO5*P8qrj4eh)WLES?_b=z8 z5J$h9Q~i#ZwW-*K9#|PaQlcMKj3C3%2H@&gh|>XTT?jfdtUf|V^`Ea1d&3`oV6jc1 zWZM4vxumq^ce4<1-o=Y)NfP$vDn-CLI?jP1`C9?Pvx_p(^U|xO&U@VZpASD63LT!Q&8bC(eg-#4mn;4V& z!2umHhkYyzGOm7J{P`XE1x`2RJOqd3tp!&&>$d+vz-n1gu(SpHp3L!i^DzxT$&#&H zE=)>#998-Eo2Zo=j74~>m_VUjVIZ!o%f{_%F)_BQ+AP}I^_FFH)@@T2_o(1QK`%Sz z*hYNkuqQxvNFe4g?EK{lsp5`8IZ14nIjnu)gCqs$9tG?-g|Q#=i}{gc0WUA>gwnZu z3T4hat;skXbYYDq#(A5tRAKPpM-{~y;$E2AH}#x1uQ=Ps|05{h;FPa!WG%KTO5EAu ze9dla%1l}BG7Wp)84^kf^9|xP&i>Y>_bJxH1+?_?wR$ApS}tN_))C+>Hto+Y=-)N- zJE!cg?ess2+5h#Ui4gLILE@Ff>OJ{plcfz!>I*oGETXraGK_5PK1;$Rkl`RcnD+S7 zxih?7P14b)7ITp&3<3|mA^W?fmbTjK9u%lQLvija+DHCkNk}u74budzYk(u z*G9yGsc#<*C@I1_86)p$$d6JynA<^9sBa^1bbX!mR@|a#e`3^sqF316auBo+646Xd zQzMk80U%-shFV_aH1}ZKtxa#oso20zv;P?vc*GqR6%*MJac%&MyZS+pa52mm5l)ks z_nxtOovBM9O5|yVJVj*(aK7xRE{{8@BP1yHJqOSqNT2L~&nlqRLDr$@f2-bR)G78# zarEc&F$v&)t%IS=gTU5gtunFclhB@Ft|WJJ!`F7gg_#_3a(1*Yi*Yi5AJvfg4djj9 zLzv>QyNM4wlj7kXU7a~9wBE}Wgf8RlRcHS+r8e~{t;S>(sQ74_*&;lA3r&{~@8R>D z0Baa^8oKHB606ALD!oU30@Ppkjh<&j)2!Ua=By5{;z#)8E$Yx~yms%hd&X_K^aP`I z3td4ah#??nk0Tjn%8pmzF40R`pEIGo-bj+#lCxz_)P8m7Sco$;%RLFSv)?uc`o}DOkqB zR11%l6hsLv2_!4Rjjv7D=`_3QOws0<&=EI6H$iLJx(8*~55GDymAX~y%0bD^mQxqN z(9)7aQX*S+BE`dJW(wP<&m-RUxq4#m;UlsWxovXFjz_Boa2De1Faa-LEUCN?&IeFO z$7Sql0!i%)sXLKQpJgp?oWLD-?0>@LFX?DdCF>5M7BWN{pX zt~V?YIvjxY+am0E_vjO~_?;gTrYO`F>5O+0blTDIz+zjDz)3*rv^s?WoShut%epzf zh5Dr^Gycb3TAH(T*k#}ZdRPs$mT76Yd=Ivxft35;E(PhsTl9T?Y%N<53l_}hSxf$* z%vS^}o@BujK?dU?H|g&5(|tpSIedq^#z?B_5tY~*Cd;X_X?u9)Wk4p8IE#~B$L*d& zq2-0V7yTX%Nxr(}f_-*BhD!ThVsJZlsWc%VAFN%SQX?@R-??D0hSh6j2OcHVpC9mT z7$qF@*!xC8<5!GQ?Eb~@pyY)7%qGrdM1naOuMAm-vgxQcJM zij}!i8nvK-xuH@EfT)k|L(bXYWI0ipz7O#z{@J{t#0NNec97OAC-KU5g5C-}#@`h? zPHm~8JlO}3_`nBTqV#WY?Hg% zBfva{1@?6uIsU=(PxSfD4RX~ijAUc_0v-_5bbF3$ZL;^Tz2=b2T^w2qPd{x*(^{dlFx z-9y|vS47humud%2zoFBap?fO`72c>P;o6X*-+Wt?FRkHOykw7h*w^FUuMh}8Hpyau zcZg29fvXBS#Ego!BPV?kE0(d%cx?uyR1*KCzAENGRPn#OKOqio(Q>ZS_!M;nlLCC5 zsdFg6;Z$n)NU!tMptf_JQkz@UzF=T#C}AuWUXwQ$py-A+hvDL`4Do2 z5(1QJKIvqP!OS@d+HLs(6wHC-*9u=K26rnbe<`iOTo;ga@^m0*^Nvdut)4diKvS~c z1Z~r&4gcJ{HQ|UQv-XIQ^B}Dk(dSuVD)iyj&UzgR(`HM87gIDz`o51%&2^d z{#xJH!g#EIaWZs4A3l-z^TpkUFpFDpruE#b$v^dn$}L~>+Im1=U$^Hyf84kiyZ3Dc zL_5&);KKSqaxwc&0I9*tXHkJUsbnQ@;t_Yc{d;ep^LZe)K)r1rbp#q07=*@%8<10tA*kOIc?TL8GYdSy6q!HgBHY_0N-qw6vP^IO9kmJ{_OKoF zvgZ%JwV1eXzlh_Ec{Jtu`S8QhYKkP-A_Z7&s9aJn5!#eG2e+TG(oMr zK9j7Sm#g{=?_798^@Tz(q`JmOEsW~qDa^#2B!Ld^BO~$pLCy1Q))J*}n<+-`P2ffuXCRH^$uz3x~XkX7>P5YOF5MvGj_ik?PV zP77E!VLxyJ-CymMM*t|q5I?e(?)r+pNjo66WP~@%#-Y5BdSqP-R+K z)<9XTAB1+B{v?-CU|Lyf?2eIAz!9?N>UeQkQjhxr_1mzw_I=%cWftVx7ylEhuI`pNXb2ovetsGf&J`llsfB*m99rW$*W%Ykb zXMjeqNZ0pCh1Z|7n=$}c1-te7D%)iJV!PsK?ayNd?N8g%2|#eTx$7NpBK;`7JJzKN z8oY#~N;?y+Fw3Y@JvL%!TaI zW%#=`k1#ZSPL0yYW2Sgp75xu54!#9p;d12>D#{_=8WhC*ykzmW;z1BaEQJb+XU%*! zpm*y9z+ErER-UVx!{6G)*6z4uZbh_kQW0Ovm!~oo(5WfU!U%nP)+SUFqSw;&a+tH$ z(cUNsSK1<5P1$yH%z;-6enuntT9#A)jD>y?)Z?Fl+jJ>)No2c4ti2kd{si8C)1VGv zHI4KxGtJpWF@D9FOamKGT?qCtAvDQdm6>u}d1!B%(Qq$P!4P9AExF1uD|hi_H0*eV z0#jZO#Nqu|ww2h^?xX00lxXM$!id;6I$q)Loq`XSY5!Q({B}%h6QMR#FQ1K!L4sk7 zspL6M5U~{YdHmW-%qn~&ob$1W3qF2G)8TKt70{i--F9VVLSu9GY~q0>OxNeXJX%Ip zZ`PAnxqj};^Mw1~Cdm9-K_^Fyz#t(XynhzDB#r5YAf7g^&+fJZ1+bu9;E|PHLJ}?C zL>Ltezez4aG*{nqR$6iNbwdE75a!Tov;YFd#sMR0iWjrvU2Vhzt^W^OZy6MalYI@t z1cJ-pE`vM49g^T4+}$C#I|;#EgS#a-!CivG;O_43`b@I__pROiZB-XFp9bjed+xb* z`Y_pagQ~l{s$afVq3(g8!nHPIk4)qZJxU36r@B;wx$B?DR#nwdmf(2o2gBfeQQK`63R-Rk%iDSR%)NbHP&zx?wqy#_+ z>wwW7xT|uDZOOhC%|Y%8-t9~uq;`7WrUke9Fe#(2d?2k_5o~Szfk0+|&Y{Pb~{iEOpAKzs=pvUuxyc`$|G8=SHQuh-`e+#9gLxMg56Iye@@PJ`Fciv$c%s zRq!);RV82aTK<=YmLkjqy5fd10S+*On1}IV+c7qjjEMCrKJ1%Fr%uxBkGI=QY&f`q zjo-`TIMF4+UeMj<_HcRR3-4#3$e<<8PZZ{SUg{c+uZV7Y3TDbjL?2}F;wl#6M};Qe ztY|^XQ&LIX7G%M`cb($}(#Cz#v1(;Rd~&n^Xc)Z=3dhq15&C0y`O**Af9JFoWVHZ6~8oO3C*xwrAqOE zrWaFecOltp26rbN;=bIYvpgx;)ynfo;2lSpepTF>@876JoQs zOGUpJ=TYih=IT>Y{~YWJml1K7+kteTh!0JJgwch=oDNCDz9wZ`r6G*0)4jWJ(H>{*vqmgS(!C#g#R{$C zZR5**Dy9&iegpsbrjS%R#JC-Ox3P-c)fE>-itvRdNBuwL_5NgX>B+KtifDZYITBif zVeQ(oCz>ZP7`IONuztVAX!&7n9lm5ARC}h4-TO;2q8UL*^f(g*+Ec)qwWrWNV?|Hj z#N_ZS@}%5~10MjOzTLkWYHN|$+DLVk+epvil{8t$G56aEXb`D3$?cywnBMNEH$VDr zrSD4z%iPBEsj2u`k`+es1EoVRvw32T>&}8pPWme}NNFW=Cyt^VcYm+;T`H2f;vJZUw_0KTt@aAY0+COQQ9_mDxSqir5l{LR@w8=YMMZK&Djp;Azk4g0j$2 zoUE%J@A~9GOAf*oTF8XqAE%x3bn@~g<(@JSjYljC#N_)(1!ao!sR8ZJkYe<4F1=!|FhK8rWR5F9B z{BC?r*@pZ?_7+{=MDX1m=MgKUT=B*7Y(F+r|8f$J7SvvI*x34#GY$Nybk){#rc^T< zI;<6L4jokp`yjbUQxd*aZs52U(!Waf9cvV`RApyp{G()J+o+JrH$yLM5dqB7j zOB@T4(VH*IJWS?YO77kSDM_g5E<9-K&bILRMt)DI?a{q;t8fDfcJ)%;b`o!jItbT9 zkx1K+Uf~@nx_<{z_>$CXNU+6!;$CsfI`Z77g2NJ9tc?$|YgzxlY*hzGH}n zyrq_O2|>3d%nqjStGSohmW)Wz%AB>c=5bEhbx}^Aygo2>xfV)Q3s1%$^jkQ zuj`9e?xql&XecC897Pt3ER5u|o8*VO+50(OIhi?8r|8RkQ7(J~`dn~@bWX6YG-dJ! zi3$6LE_@Zx$kmh1)U3A12wSUijTxuedmoR|-359Q(ys!o6#{a`-Ysit39*4tG(V zY%|xqXasM6r^2IK2*+)9Qh=_~t4aHk%h6W(F|*=-lYZ+H9>!jB#LjfdT2DzXt=@RL z++Hiyj5^^E-i zX@xTKtCjEN>Gf9%rt?8$|5BFmj18HpiPEPXG9gnqyKRNZlrk&pT;+?&oe8sHTdHZX zD6quSvz0sj<;<2Di#*dWc4opeH_poO`{N$*&Q-up(nPyK35R4V_@ht# zrZcNoppQLRzH~V0Q#4Zx?M|v|W=bBFowp|FB^>GfIJzJ36*8`stic`0`|ge#GD=2H zSzusj_-Vb%(T3Djem;L;Yd|Sm3@J@q0XtwywBss*48spOtj+Ccw=nbyFH30jV>ymT zHUfOLZII1FvQ;Q6I;+nOvfD`ItgQ#WNo^EqW9>}pvBGkdRBg<^RM7sr0f-i0`3HVc zkOY+>je5usUd$co-QCWvu2faopYJ7VCAN0nqhowv6?$Dz5$4^x6ww()Q}+9vlkJ8j zK4QpQ!Djp2{W7w*CUr!M&9@EYeaq48zK?R%-5q%lYXoNJ1iXc-7nN3NGA9xU8g`g{ z+kMSi!HmyADfEt#WI%+aL6W@iWmQFe>v>U*cOdT(8RptrocEDX?n7XWnOzt-qN|mW zt(8CdZAH#(Lv3o*TR?s5*D+n;!f8?Pk7l@#EkQcDB6Enqor#JGtyEA#MDq78yw(-z zteWT6mf8NwEaR1G3L?DgkRl(edY;9jktJvEg|lni)Y zi4TwF$Z3*j&-JwwAf+v_?2$#toTv%Ov+&2fk3w24*z1#-ob~H7zO(rGPlL8)g?W)3 z)b1Tjt6=S)`+`yH-S{|Qrg;T_w`){_u(AK8sRXe@2(Z#%(*O8#LLtCCc5sQXfFx}G##`?{N z5<{D5>IG$@sP1!~iKcspA+(xUE>D&c>&5voZ}mB#f0%Q{^Ej1#Fl5Y&FxuB~%g+7z zTRnWeGY?wN55lE|y@FQ(dPqyJq`6I$J(3_=jSu&`gl(Z-!HWf+6^dn_aIsX%sUx&h z>m3c=9h1b?&Ay@3pXH9u!V)D)N?waxknY(qksqKl+?{6L<`fBquFX>g3R0MTNs|-L z*!Rj}LaV2=O{>{7pjgOL~ zkRY`nIyUJ~k1H#%SQyWco-2jLsQ))^`Bxxb5NhHX@D3*2d=_%Xjz&eV5uOa$gZxg4&*c`kWY={ zMKe>0@KRisrnuYI=}GRe3}?;w9?>NmBUzwZN!Wu#e!c^dGRpp>$kc0_(2|Q}Cy2LH zsT`3L<)+< zotB+`mpB)SqZSX>!IUzjqX*8sL(g?~94p%`#%z*6Ln}^g5AKV$4djCAWM@ zLj`$L|M*8}@uw9sKiVH7Yobq5=1-^r10RtZd`5EmsP9e?!X(_Kh`m25{+e22J8Fp6 z?$|A28n-@6jy!l@n11G>&>?<)mppnn`V&Us?d@+fx4oQp5nHH%{qLWvSoq-`A2SD@ zVno`b32ChD0FPZ%(%z2HxdX50JYFIFw}h`UB*Gpf;t`{f8BL1HgnqasV_@yn_Gmz3 z%Y%j$UT*O#H*5de?Y&|19)=sv=lAf1i6E50@bs^bk^04j8c{52E~^)H{a$rv)Z2)m zHb;a%XNIpIJV3_#!*|RnUMaP|j#p|zNCQ5}b+ab~NhI!3DCXm~m_F^7x-qwiaF*^( z{bu|9ND(smsX#6XLMfRJZ8fpOf60Qc^26!mx19+S+T+zvNJ+Ox8KYRqH(Y3npxsU z)l=K~&XNb0bZf`GYpSAc?B5@jF{))|EW2xDlR^jA{wS};VSBXyEjqwwQ|42aPTfXL1S z`dVR2{@(33hsz?TzDHdF5UO%PGPV@(0`MPaS@BwEmrB5_T<6pbT4m+=S(I$N|rCPi$w;g1Q^?lT>Q{Y*z=mt$|Kp+Tr8a*ZOJeHCq zY>ZtJf4ClBYl=Bzw7#Ws!3CeUh+TG+LO+erlj-v)#(kpr)5wZ-HO6_Uo(jXdmuJs4 zK8$$Ak+H5;St^lcD<5n6W}ozF3x$|%CyeBuam1I89d>%J`T?|=tm(_H=;nw!QAPz? zsv7WfyI%s3nz^v?y=xP--sE1UfHk-3M10rgnhW|dliw<(xDiFTb=3QHq#vln(y)Y& zrYi45ifjL)F}-{$ee?g*0thILf>>zv15N3sG7K9&betx$AJg5*HU4clM@i}cvE13t zZdtJ5gGZ5izDTm00D=PG9#GOBP0EDPXy|-)DCG@;reB!25<9wcTaej0aVk|*AGc*l zf4EUd0sUld8FLs2rIMD+Am;zSoR4gf9kZ?5X98N_%ead9w=U43VSiNcQQ&xsvz-%s zkO}upFM)7S3H@CqyN+s$lr_<&<5IQN7kA^h0b)_D^glw&;{WH1)liDC`pnNP#H}VF zeF+9l5_*Uem;>+66I{GR>AU-he)yx&tRa45EMt8oa6~!jgZ}JY`cOwzvAuQfsOK9`TD0-J4#rS zqYsrRM(i7+VH{LpotzbuCa6(?|8H}l2#^eL<+yYMQE_OGsM{JKcLaN=!B52<(9tfO z1?iG8or5-ft z5|DWo{=@nZf+5F+aY8YUN(azH3lIzzDfUn{X$qgiz@z&}aNP_O6pfRl9_X(ZjRSDD z29ME%*Lwo-0Uv&Um4oP&a;|nLx{MFy=wc&%#OkXzsHuO}__h7{w|uz z?LYF$_pv$d3v1zBEg!h!{YZ$?EWY*SAa;5hid@5QZm;F0mt9oI;&4hV)u^~>##nT3 zx?jP{82QT8!UBUVuT>)H1%k~(sMY7#FUt=nk zCm!`aJ%%B93YPpJ5L3ZZ4uLX>Jn%H?3vj2 z^R+>jQeN(26-*a*$!B|4we?IGhWnA6E2oGjnT%h##e?jH;lT;eDbr$sE4}|KsVb=g zDa|JoXGPq|0Q-#~tx>%t8dv(Yi+%=724fyS_1T1Kf`8?I*#4`n8TjS zzVfL|)V9x?WvbHkL{0q^qvUOcv~7w#u0S=&Mot`}e6==yrW`o##P{f+^0;vgAK=Cu z;!m-*j7eGIkKEN-#D5|fDyoBY#gr=1eqI%M?iQxwkUCAUnz&PrtBC|Qfif|P%amL~ksOHwwO>sr zxl_M0%Y`n_g0OsXJOl74dH4a^^6mJ>O@D>fP@6Fh%|j7 zsi70ed-#GS{EwS{m4kJ@NX;*}e20zKjIk@Yp(@;XLtH}4N~pcxEfC$XG9MypdY5=B zyV08|YckfR`q*o|rfBEwX-bv22K+i#z+68c;1AU%bAs#Sf1Wg6Mp=vKM>>6AjEX<&et5W2_b@^508?$4>7`dctT5g@CJ zhb}&^T6BDS4{__+L+pUzj~&OFk%9xz)$^IqXVfqhMrQE4s%5>RD{Q-l6rIhZ#pKN= zWU*q}>Cd#Oa4RVf=ED%d)L7RCyxyDn(Wd1ReM8DJXk)t2&Tk|*Iqx{hhGHZ>{s_er z-8|KsZDF<7Wg{*stA^Jv6k?aXzM)~FxR4Z}IqACWVp27DGV9M1Ph^VL?_n4>F|9Fk zH0VW^s1Gwd(OJKqCH_+m2ote~+2cIz76V1gVy!P_wLd=bz7Nxi6^G0s>268KP#q?A zw?g~TWZ<>PkBkVw>|M9a<*nRs8!#1o5U$8fb+g58EXd?#(bEzr-!2@VaT{ORk>L(; z<^h+FJN=J#I1DsMXhF4No-3#XQ!M~Mw(K2qTN*#2S(s-9*ynxmYBI_fTd)Ycv9~+R^d;h1X*!6Ck}6CObL_|(B5%Ew)|6USI@k|uc^V5vTa1ZpLJ z@E|7kjO>3SH!(=Fi7%yhCBAdpcKKJ;DtvgAYy$oAM6m!q#F~N zyL&so2g~^jcpdBceEXkjN*;`DEfDztBix%z-2tQPGDn9rat^SnKgLdhO?`t^K>i{Ii!ukDl_|Lggm&U5dy!0Zso!F`k0>Z8=dG zYzw}gt%q|p-}56)#?gcUrGl*xVE<}k&2I&`KMzIb;{aBfWBaA5?>d+1v15vty_N-~ z7UA@9KUL{qica{(_a89{Cwa+1vUGc^0Qaht`949&BbooPLc{@BLSmagyWYi$(1Xs_ z^+lpdp>)emO$5`RJZr?%V~VI3on&*rw!m2M@wv$i1SbB>IG3S9?6R<2+>Y6OYig3a zZy4H=@NU48`%`%={ETPHl{DW%dl`;n{Bbhpj`51Mu_!XDS9g7Vo&xESukOi*rWEq!V7QoyVo4d zWMaw3AbpsvLL&?Hg))=BEK9~$GHrE?y)qx?>)(W9{rU(}E9tPgCIRC_`e1|oF8xJU z2d$V+Zz4ro(~`HcC{yv|taY16E4K~eR%>0(P*R|93>8WjC{{fb-V~3Qa~ve5>cy|} z@;R`t35z)bjt>WN<6fb+zv8`PY=q9xX~MF$nyyy)>9GbivD@nWB}s|z&&!ZxF9hJ?j0$$iW_ zO`LwzWNYwxUM+tmVKsM}a7bm|>|+++yQZwj-^B=+gw|X>wRM|Am_z{I*}VDOL`l4T`WTHo2*f@q z-0$R&)5w>ladKAbTSk<@Dh9m~^%19lMgy2{J25Jwp%oo5W4}`w$yIA_C&`d-PNmcX z`RL#W3bjO9zTh;{+2-AcUyXR$8JlThK)bEKkHc-K{QDLnBB_=G?!2y3@7FoiP1_U=rF&n8 zdYK1wpMv!x880k+vcp9Kq%E6ti+qo9;I~?D^%6+ATQ4iHX=H`Nc-DkzFTY!KC7$>)mB<3TUP=?J=m!k`b01 z*OJ(0|pHviEL|0*Fyg2Q(MN?~$0VBo$+nh(sLtu2eOYjW=;v zB^NELq1?R(y^Kqe%^=9?AuMrqdEyxJL){M6%3HBv$NkJ+ntXcBx`I7*?frDKy*_4Jkv9_rJkD_0glyJ`cJ7${ zev2yFnVH*L82e{W>8TYJeP53SGX?x5R}RNc(;~op(dUbBl-P%fquY$+|LGja0xT#* z)|G#?f{4%{4ks$(Eyu)Iv56YD-}1P|JCf-Bbn3v~y5xaozIFahb>BoRtqc2lILeIB z@(*ZrmWV8oqT80SE%+8R-h)~dB=jsgd%UtI@gg8-{?3gBTz^Ci7fKD+KJ$)SE>P!$ z2UTJUeEf%k_Y=+nDPB-~z!O_-eJ(Ya=EnwNADAN{uuDRjm?A$p{FjA2e&Z5VW6j&)|{9kW7T3E`w_FWhxn!cNg}FR8dW<{GKK^-fq0;H?K6xu$WL7@_U z;Zgc%)c#4ngIR3b2#7RIkJA$1CBr|J_l~t#_3Yuorey1<_#nvb^{OvGe0xc7DM}nr zADz%If6;niInNI0d+X`hr~m;J1IXOX>2-$75rhT2v8#F5oHnNM{-HD``M7(?tDw#B zA!YrhR=N7l;05RLSINR;RZFg>C(Gei>N~djC#NR}qcr1=y=0e@pV#5G+L*9MGE+x0`3HcsR9)F+!Nav5-o<*Y;&wFr-tpD(0&^x} zCDu3Ge0f~5z+-jFYaEF_0gyD!9Bt}d0VP>13r2@TgEPK8HvhIozKRuLg-;iPGXYkW zWMrTR=&J!AemC#8dn^={n)7}X{ekj~Ua}Bc^lLgHzwheiVTvA+F{%pLQvaD^BZsA| zbXjmcY%d2%OfMc5Nw^}(de`TK;;gM~9OUN<9Boxh%h%-SGFI6EEb;Idvi=p z>h_nX)B^9k$HR;}GoSsHNOn zTAw0-@SDfXCPww;G!4u<0nAFmWr>1`Mo33wHzSki4%!KDjD}$b7W3n2Y`xp%F8?ZO zTukgOI@^LYyn3)?OZ%e0iMR?MJ>99G$C5SvIlZqlZjUS`kK6t!;#uRvxn9+2s2&Eq zU;s5K)a0!8(`QWM~ULZb#|oxS<>ees(guGu-`e~KX`JWBcYjQ!nXY|OhembI$$6=(s*aPgp_ zjad7IclX3&u2^Tyj66=&gP5p5ep}@KnKdB&x_5Q1DdV_YcZO}8-r*Z;hms5m(>R9|J z30Bz2I3iyA^*e{j>>rG}VT_#CsoqQE(6|9=6cM4~vu#4{o;CRqHYbpKUndV*6&6=R z;urOuc%4>dyFWe@t!K%rPTTveBM*68>||)X6jKWDpoibw&-X=b+Fwey)YnszObsEu znQS%yJQw}4YuhoRdNj20`zJsGo3CjR1>4^;y2-1>uKgn1+wIw`tVSQ^JQdKV&(BgJ zWGgFQQTZ<}vE zytX_Ri#j}rf*5SwPGpp`G6{}!Mlt>)^nqY^i0hx|=aHf39O50n!}F9Gxm7rSrbAeM zhEim#Z|1J6D|%tDTotO0y6=LNCMugg?o@>fOyB!qoNs5Yx$)W|*_Wkf6@Ak_+CF&Q zn6J@Dr;E3~y>PlN(~hkY4IO%1B7=4Vp!p$r3(PJX+XNr@q?7FED2DS^*8Ahb=&uYGV=w<9=Y&h^abenU`{R z=#t3(&RMBUqWq0e3%jLoBU2>aHeJHGy1@3q=9XqCEBUH&d%1(hFt}a?mfDT;G1tYe z9~lTc5a{YyxF60D`RUj9wxqlr87$h+?B@GyyF8`yU*gU-Eb*63=$(^lNpxS?)40=? zOxs=vJWkpcw>DA#2Ao`gGKs<5FW5UZGr!hO+JS@#&}M>`<(NNOr<+m6CvjM^UcnrI zx2q&8UUE5n`gJn@>GxthAy|^&sqdA^i_pUX9ch0N1CJ~~W2!@=O4cm;K2|EYH?EEx z=kuarb`Qf$tluM){cfU~U$MY44PHFu$+5LcnVWXA*O!{-WhG5~;e`dYVi<~<%;8KY zuNwo+oDXJ2N4g$Yv=dU4v;+uqKzDmaR?nPT%7oL+nq7Zc!BU?lfz&p$r>`QJrOXOz zU@s~ttxnvsK9&SYlIvo)UtdB&(S>#2E0_4P*Fiz<8lpfBj1X=BaXX;xZqZW>yBuof zrs`pHU~E0k?REKdndQyhvZ?ct0;$!h*~#Nq zLOS|+l3p^YK+02yP~4OlUng;7VQfA1;H}8%KnHc`+wJh}0f2Lx_ypH9ob)zl6a>TZ zl6D4RErcAoe>Qn?vZ>!(%5As`aQ~&$B=A#2@QkaLLh%&~gh;Awnnrg&LmHf?FL(HI z=F*zfaHddhR~*~Z0@_AeXASB8oZG*z9so3AmeIQE1|H6xZE+-ZIV`YNzjc zYuc{0fVdrv#*e5ot_0Ku>7Rb63CeFQb$|CW`PMEwC;*Db?3BE9JH?&h(8`fs9t#!VXk-`j`nMm-G zr-IV-;;UEIqkf=qdHZ9|+m(mZRNv@!e(J*xuV0*$gVq&BrPs3auT?j|!c@qLgOdp* zIxRGI$(C2=%U;fBf12`omTNpNQaDZ?I4iGXZuP8j*(ZWg||kbj{U<3kTBiYbgbtQ=Q)Pw%El{u){d(jX*_^!i53q0k1NO_aQ4}=L80df^Ff>8p zUVP(<1CbBX7($XpHLf2(?{V3;XUG$pGmJ7ZBOZGx!^BbQn}SuR)_eUcQ-fV#p4{xJ z!buWBHu@6LqcQQ=)kk%f-kQQ09%cq=3d6qL;m(DJYdPH_77cVxHsss9|wxN1qe5U{#@!F{D5Z~ ze})GQ1*>}R3LHq*BDgDW7OLJC@u_8@(WRBU5ICW&;!xiGPvGZ-0RQg=hTiiyC zU+W&o;MSv@;tFd*e72aE07H^DfPZ_zs71$kZrys8hL1`eME2X*`o zw6gE*YZPu@yzpsB0MIG&zXZ-(8`}x#p(aTPZ)@s0&1IAU}Q=`YJCI2JzHz!DhM2Z5Fu9kH;H`;#KOa>wt{bo!C?T3(_4DL)5 ze#>E+lJR_xH`dUiD|K*n91kS~tLH4_hOH<#JB1FG2Qs;j>vRNek6tg!JS;PQf=g@m z%C+`7Xm>n*j6F48p!~8r0eXt0nS4zixM)JnK#)z|({0lr01D`~`k{IXs|&vi)8df0 zxY+jA2>!F90IKZOU{8g?KN(;#E7TIV<--DJZRP5^I7x+}t+wy+iD0y0@3{8)^t&;I z|8&Rbi+^C~cs4o@0z+qH_IC-gGgr_UMAVP$g>QtKkle@moJmPWBuv~YMQ!)*znc_P@Bzj<%1 z{-+EnLJxhwM@Jv`jF#EuJmaP;nc`f{rdy!QqC9W|AQ;@D8Tl(!6%lsfgGKZtJTZ$b zw93_Kc&mM0exB{V%*b4MUbuA7Ky72;BY)!BV`BpF-CTN6goSY%>F?CPB>fTQ>ULe3 zq#h|%>etaKrIjn~=M%H+NceINxu4wpRN?AI@0fnpI~n2shs%W^RRbiKwFT-lhgeMm zd3O9oN@i^s>XJgGIp*4kwlU4%Z&Hwf#Y4^`K93z*qJ@Hj&Z*9cb?9=w`hH8?%LbLc z_E|-yt{?s2STP{Bk{B!)ylmHMdcfzG%uIW~Uj&Kpl+d6fzhwV(UriyWv#;aHx+TRB z%zw4(K#rC~NChd;%ywI9qe^oihW2cYa8xgt-%3%|FRfUm3|;j;h#Icj5+6e6ox0uH z^J~FJe1~&>?&V8uwt`-D{`Ig=9^g({sVamf!|P4*gY`ujsO1&|en&9o#m0ITy%w`K z`$1I9ktvz^v;6{+8AEtI;p(h#Db&{PSM^Cr^$e5sz!%^C301-W{zgZ z&J@i*11DPul7CC%d4zlgJYm2&MrgsNddEP`jc#2zs`tE}l7bZC4~@!?Z2K{%u&0>p zr*&;utyJ(dCN!c-W|)*#a+PxGvF68?pE=`zcy(d@maqM>h3~O3-l>dpudv=XV7|i0 z56NwTaS6DqGL$TvXf4+Azk#7H>m-|`LXCDT```fcT~8@_SpoB&9#K6_!h2lV5|}?% zNApUqNC@qi=@Qzhl2b7{B`d- zdW&)JJM0eyu0v$2rtl9uBhzCDy`))>g(B8%k1B8(ZoB=vp6A}-3}9#`F16f`4cLo*63*7 zKvq#$W>6h0i5=A5tL3?RX`d*Y;HUz~d!GYm&TEN}*$uptOi3@T=N9eawdtF^1oiXP z?*6uzA?xR_|8ts8&%jV>jdxAGgX1wG#R#PfC@Nl)UM$TIeKWN=w>1bYR~}NYH#;1! zQSId%tc4k(gu}b54mssMHv(tJ7!+~SVY+R%g+>`HTizTSm?)v)0B47eB!NWMwqAG83`s*M=Xi3s0mYNg}<8O$R8M$`LFADdjFN zUu;(8*_A!l${8ZD;;*GCiH&cC^K;3bTS60#h0H>-R0Q@%#3cudy_ksk{LYHe3rr)12)b5iBKyBfIz>kgis3tgnTXl{fdnc(VTn7%dnG7AWY<$lQrk# z#njF;!c3zh@$xBmipU5HUBR}OU6T&zbK*BxNP@A|7K$Tsv=-z!7{lDOcZ9QkK?5>) zu|TaHf&3Du-$A!e?$5%*J_!fIgz9JhKJg0tTz6VtGp^kosR^fAo)v#cNddk7qzW#wDq;*7xB;$^#k=dMQe z?|b}*zu?h8lG_pO{zMQm>2C)!n*OFKtifzpc=S40M;dcihr%Us?wcjrH*_iev|J3s z9e>#$jf?#FJoUKq`KPq(z5*A&J`;`_R#vcL@GzYIj@|`SJfj`~0~8(A$eMw* zYrw`>U6F|v4=rn;PE&4me1RP8LhRk3^N_}^d%(+*kFi}DVfoD)87x!^sm-h=LhxpeiU8H^(NbUtOP4u zwDEuUrb*2#kig8&kPSJcEQ$Xd@9ajUSC#ZKN5z*LQW_sR5LThtw|g>v=_PNQew!4$ z+xWx%VEp`~Hu9Jg{3a1m>-kDFsAKbjz=y0NvFN#ilJ(S5f#p_;%mv89^ktS&FizN9 zZk6j2LA+2ZQ^mykhzEq(<@dbIlB4N-3Et*YveHAs|OXqZI8jp}h4 zi;a{l`+GB$TI2VYC(~B=K4-6BEtO|s3u-MDAN60#YZU4KlrIX&?&j$&AQfx&9NhLf z$@k==ti0K=a}lDqQ(OCa(;=6baabAnpTO%6vyt+LAwL~lEI|&%1|dS1a-X4h4(Gjk zf2+gK{3G;FZqK94k1Sbt7wLm;%0-{Zv{aMR7A`c`!*#huC*FL1H`}64ebej8`Ut9Z)3eP@ zCR|Hf8&ZZU=_~Z><-35scwLm>83!q>eHrd3zLWnAmC04q9Gf&a3HHH|jD)d$#QNSs zf6x9}LO7NyeQ~Wp;SCd&p0pP>WLm9We)_5dmp+?fF8tY8KU-Ev7)?5InnkB$qg2!3 zJYq}HrDiZ3hur#eW5YRXMsZfl1p4ZM_2!{W9v!AC)U+CNV;+-BZXccF-^4#NroA*v zYf-z8^V$#`bxn@ZgqLtKRt;e~!wfy5_bLMI$SEYE{O@vBh}WN>?^$LTlM))ElHx3Q z$=mKBM>llc>rP8*+CEtJpva}XJ1zlJqY!gDu@zh;HRSHqKPxajBuqC{7a8g8&~CwzIW?jhoSoh(YM~~D z`zMYjY>yO;x2Jm9Lp%nfW~X_yiMZcEn6Iix_}51hOjC5!C3)T<*&D4iNpHXU7lNh| z0yEVpi$vGmr`@2lb|K26e^{7Jk>~~Co4wjl?&dbeGb8r3-mn#`D#ipDSJK1;=hZuE zwHiW+#;1;WlmNJjx79;#Y$MzN^LpFgVM+yzi4_8lF#C>!xZ${uZqela4oz0UsMG&f%!vcmcN9EuDtY5ODYI=JJ4k2X1jchQuqOIeSkdN z)Wxy0quc3skB}BdIsuNv;O2sX80cD;Z(gY!C$_eIWC7T0jk?AL7jN3@67qXn5G<&( zU{C5^VX`sZ!19i_srB6LFvXG<^`&Feglh!E1It2fGSr{N4DjU&#{)QgmXVZAlqQIV zJ2{4(FIdLfNY0d;3?`wyfct8^yBHp_!45P4n2EXk#W6#HX>|8`7nrv_%u-~72#pk57DB~1%)1+ZrL}47 zSks4r$c{^?NDkb9feJL5V+pInyQ6sxuh!x62F;+ZZ=Z~dXw59%$%)X^TB87F+*52G z_375Ifdc!F0$k#gm# z$e}w}WEWy4=>I~xvLL}l2Eaku-1RE$EFIA9g=x+7#aSTtr_v_a+jq$ATZL;oe4SS~ ziD(6(Whc4)1|zG}8&;Px@8EK1=MTJ!e*$s{d}@%5;kc(`%)JiK%sq^Kq{f>sZ+x4$ zGs=1IJwHw;pZap*0~E336VO-2da1|{nWaQoAy$7&^8R=7dDY((_MME#DcyAJp8QD? z4U+vCikrd(L{{W7pekr?R{PjOgLUV#H{(ADnLQ@N#Vz&BlR{kF6p+VPyel~Rt#6VP zHa|nco9|Pw4O+Ip;1&zR@IFi$lr6$-fvGo(QIIWS`9N?2oIrOwGmP;iM`e34yU?h3 z{sG*QvhH|lLt|?vo&?6}A31|z*s%l{k0=-c<4BaXBRspk3sX=ljk-Ni9s%)e`ww&O z)+%hfd1X{ZTRXRuK=kFuZ?!fvm?CeaQn1Qy5=EyiZcW$n!qp7b*UYiBCE9pV+F$Ymoh+uv}voec^V1`>C#X;wp`y&c{ACl z3DRl|d9%MV9yK-NWYd{h($Q3IHttqFmetFmY=53bE$EAQcmK3J@2NP!F#;|A|JZs9 zsHobme;9_LMH-|#6;M*TyF(g5>F&-!l$P#p>Fyd3B&DT0q`Uh&=zTxW`~Kf=t+N(u z0RuDFxvstUFZMn|aL zi@x;*?V6|E+=or7*~!@90Z%_&hsS{byRRd~fXbH|Z=Mp+)w{v$URN>`1%?Bl7}l znZhwn@^W3mOvYo`>m!c%`{QmKm2}q>sDP(n&1mPh=W6p=>|a8P{e};!0dNdt61vPs zMY0F@XLoSZM1JIwf}7{fbc(kxa4HNw;3pcexD+U;z{Zp zadjz?2Gl}bkgpU@b3R$}+{9saxc5(iW`6n5hcjw}=~vAv-wWE65f@By@14rJ{}1yi z)e4Y4GOgwO!>ItA`vT?7w%op^$}GOyee!pj*9gGrj|_3DQ=sa|ix+UR_)m zV>bjPNM5sX!f?#vdsudOk(xOh%$IF9#hBp~9cz54KNMDCu#-k^_3x0~_FZp%hMpm? z_IDuzH4{`2L!%MbbrwsuIbWT+Ua&|k6|z@S`iudT>mAMBoxOiH#5+|~BOhpRO^ zhu(Tp@5VeCV)~KpC^5TqysWIL8kfWdT>G6ngm&!@(oZy`!2Mr^_CFrHw|V--|9v$8 z@t`jaq5esb5kk7F5K#}6{qY5xwPVAmXAz`U6RE=MLAb?FaMepS7K>6m)SfannR9u? z@_@kOV;olQW4;{GV@xTacNXUAS^j>S%`DPf=(^civ@gQ?z$n?mo(>Ob3v9JeFW%*5 zKK3rBIE@>zSJw)fue>r(mDtFY%hk-0X52kB-}>WXw_b{;Dqr2V#v@km`-kYNm7q(^ zo%M%|6JL2$L0EYLwW18~c2R9vk6&LdjL!hJ`$~d*rG>(0j9v+%%_-fQAK^{;qqUY0 zs3tPCKLkIgU!tkB=rTozbRIy^Jg+mMgmw!BSYn zPLja4bOt#i1dTMze^`L1DrSLl$((EuZ{^(UvCDy*jcC;4{PAV7k72+w93ol(6TgMD z3BtkUYX`2#U!9$Qyp7#DC^ef^*Q3r4Ju>@$WfCFK5U^JrOrGd_h?<BF!os$XR&Ol!Jq@V2XTAD<9> z@J{AbLf#2T+%Xj6G0bbKce60Wo04O@u+sn+^^o5rcSP-Yx9&Ss?)v#JB>XH+TzW1X z;r*X+;WZ{3{&x2yJ8yb)Za2}%evI**Ae>iN5OGneCb5z(E+6jr#{$(#-=h{(=DPj# z*NG|IK`|97bJ@pS9Q~w9QegWVfB}vw_s2Qktin~UGCEnpjmQ16QKK02dL`|LX`(90 z|M7$PfujLFD7MRA=jnDCP*$kYI*0UASH_qpe`T4IeS6BNz~ObeORdt;HrX`y9yX*e z0H5g&%vom>4kW!K9%fq1Oq2@@4n!ssWHTeQ@Uyd}ET}@(CE8Pniy3ucdKS~yKu4!3 z>%k6NmcZRX>~Pw2d^_ooHu(O&ykB7isQY>bH!9%}5IBk# zD-V-L)At%jWI-L|nf|leCt6nkhr=4K+P@xqkr!ooVg&CXGl3Y+SA?0(+9!I?7_XS1 zPD+_Mh9niEK`{EPc9@7Us8^tXV&WZad2&b+bAQI^P2{mz*wK5^YKIX3lP5b8i*GC>y{3j9u!BJ zNPej$rq~7}85emSG8yJeJgtQgzkF@A4;lG!Rn@4`0>5z62Fp?ffYZ5*p+qryIpEhU ziCEA&F%lx-G-ehu6~`&2a6S+OmBo88k{t+L3C6Chlm}catQz0^iY)on9P$aX7Dm4r zk`2QmC@NxIUytR4z@?~zXMe$KPrV*fxN#uf{X_0>f!d{vmhiAvrWT5DqHD@24OGV2 zD>V9QZ|oa;)_#!#Lp{*pMO!yo@tRsZiR5ilJX{qFcUQF4Z50zMV%Wd;cORxpKk3tLw zL>i@4T&kV;h2wUCXYA!h8X`}Xu7dGe!51yJ%yojNVdj3^DO{4*`o0|3l$AMA{WlNKp51+aT;y=jz8?c< z1c>AYAyfKz*sWrqB(M%wq6QD@%Y<5H%jUiVlAC^o5=G9#te5wl+>m2*&{V%E*8h>^ zq@cb4h`SyRdwPAY3P`Xl_6$iMB&je5H$wDIzcxWlePe$ht!2QaH(;3Dvd~n@$r8;& zMBKprcvWh;ZDGReGXYzqD5dadb_(>%Jrn?2=4T*-=G@}5ZuT*;;vFGLX;upgv49UKGHT(Wc&>Vl2|o z=uOZH?EMu$r~c~fa8`Nf^_Y(GNUyq>a^4 zWVO3U;u-jy!#Sm0hb#8($}omyjF(V(W858z?+ZuxZc5)$70yk%jmHWNM)x=ur~uzwFj99} zv-O=LM70}qLXWf_X~t7?MP55FJEkP-SbA*J{3p*`{Dym;<|Le zE${*c%H?#6%~nom%v8a*#&3LAR|LyhwlwDpl$$Gh`jb-L8uNif8?F%uZF&~gp&+D} zF?MduwA2Rn61yq`J3950hRA(ABB3atN9ho$w2>R1q$r-=4B>4V?bePG{{n+){S?33 z9vB{4-f<~K{uHT@Pw~>ASam9xaLP?{+Xa-frjGFe@2A>5iU}~!eYL3E$5oh~jmb#> ze+-2FhLflnz3+kCXZH!vi_B3`2JtXX<|o0i3YK2vl4>rVU3Dpr{YB+eeO2y zj9;b2&%5$@uiDJ1z^wI=r*BN(C%ENu$@WBi?d$stb9N$wHG2wDO8%CAh z&HMckl%7tFEzjC#Wa?fmQv{ExVb%|}%`a3CNHr{rlBP{lXt%zwXW;`6?au`o!&9q4 z0`=V2k;XP+lLWN8kj`!}Uz`;@F6@fgRLe*(^2(bhNUO}je|G}8L)9|d`JCqZzSIJ; zjREtoEqh#jwp(4#p2bx(JA70DUhE^hZRrTYSp>ILkIgkLk$=dIHY`n-Zrr!^`XW({ zTHRXJk*$y(_^&$dHP%HVwignjFz0Enc(`dhg?;5zsg|^V27#O%HbSZLTm^v{dRh#Z5KM8zjBwR=U39p^kB44U@CCHT1TBLI*?QOF+?@)=3YvaI~ z`zfXT??I^)CZe&8A+!ZWmMO7qYMHxWc+|*BIf8ZHBVi+YHKgs+P01Ut>qzzf|5c+# zQ2_dZQA4!`RuC^JIXg#5f7iUb`bc`%MWWE)_}^Io$1ckUqY^S$yE8=THd8Q%bqMVJ zVQI!*>gt;rTnR1iFGo_8^#_$YYh1N$97>8bN5^zX&>|uw;=RWyMKY~?sX(90({^7v z%QIX*mM}$V3w02;k#6g>mSXW|-oHo^nv3bZCsfWvw_kHG(~87{g$mr(X5YS4~BnB#Hpgc@PaKxvcIEW;aMyDz?j(6y`x= zOj`oz5)isV(&h8SBbux0j$%+188asgCeL72B+bRoQfty}jO3Jjo*NU0lCAJcH<7+% z%t07^Y&;k2H_m+w|B4l3Uo}x zk7A_*6w@zuzY94w1}mTnLf{O47|0^JSMrs7?J^^*BCO<>& z(E0qHiHA@l{}7XlcWV(Su8@mR%}TP=&(H(0%%GfpM|x$I zwMpM{=y^+=+D2-20=Yj{Z2~bB&nAqf@F07!YK%?wc`AulyG$OMWZF75W8M6d{>Wd0 zdQ678nk2&I5G`dA2%BA!ypAiXj|7)d8;5SS#SRb^u6!jJ{VX)_8}vzOXqx`(l_GCK zO1-+Zz4QF_Zj^)mEA0i+w{LRrstExq2~>849aWwM_-&~(jMsJp^4 z9BS%6n>mk^U}>sbNN2facl>mPeaT`in$FU6cjL7kwKZOK>QMdj{MB4jd1rx8*eMc3 zY3&V=$*Cj%N0Pw+Er=hnImHu*e{o{vzO9?VKlFS&fIqk;)~xz|n`GmQOP`GT;`KY) zLf^W{bK2O7$$Yv*F#G)Z7aiML_CB&#c+?+Uw;~o!Pny}iXO~^mcDboMQtEWy(gA|q zX}~X_SR-jiDOmytj*N7qXG!Ku7ad$M@&zKY3*Z)UM8_y|C4|KDT+4{zr*{Ti7{nk% z8olqoZx7fw;%i;3)!x_iu7?(o;J$dI_IGQ2xOK0sUY_1gb&eW1k6sbUakY@)YwpcZ z=o$NuSqeiNgedSx97rc8L>p|Dl;`1Ww$pfaeL^FFFqV$(av+CenJVZ&{pLo=Kd5!s zX6bA7<9$Nn?9NpxcNPh`4&bc{atEbK23;pa(mlPS93T{hjN*U&P7AbG*%gb)CQ)() zE?8fsMS~7_Torew1*ozHGZNH}3A1SJ-}!$xe;k1a2DmFWRZTAdkA{t302+6m)4?|P zrCC)GmW=tsuMi{LqOie+XP-XoC++J z>4y?-F4wYnCS?UY)Lew|Ud5%}&hs~v{>kmX9@!=>`zC*hgK0Ww&YGC@kid5j5g||T z;qNt#`Ie#6Mm{G#&qTg?b)fAu5Pfa0&N=~J8M&#|5E1K)LZlFAwy`^a!ct`qZsaro zoj{S=Pb`D>d-HTUAdJ;u0%w2rJ?7Uy@W)8M@!b2e2cazDH*O-rgeSI-AatpK7tJ_+ zi%p3)NOp$ewj7#xGN4<@Fz95Np1j_=+58LKhr4lr_VL1e>xT91wa{2{Z}kA-!1@HY z*0G>x%z&YHnJV#%IoKEVNNyV%me*ZF(0$gKxKyTDgbDAtwn_e4Nck*XUT;mgxNQsO zh|~h1^l@V!)Eu*CMr&yBWHIJa9~EA5`Nh+yq1q~ zwt0N$>bd@eXm4@tK%CA;KoEi>oIjRl|1kYVH&Sx-G#I;K*+Z(S0<}P4rE_~cco=^; z6(W(54gBdY1W11_Twr|P@ zKFS}(7Iyy3eB@m5FVMxy#qV_m>8#G_yYV_o3N{KOt0_=KU)YF<%X?za1(-A2 z4`PNz{JqJmDyk>@zG%H0hDA*fUzpa`mY+lIvd1xp*YkH|tsJdvKnY z8shk7AC9?klxBt;RI3$#IlDfkON&Ua?}76;C_8lpen~xRX4HHft0(UMW)~aYG@-Ha zs*zuR#CoH6_@p6o?quY&Aj*#g>@aT!*Y9;IZ}3Rh<^<*W8=m^GfPJkcRsw{D<#1@a zp^$jew4lSRuBEwRripSjOYOn7Vo$Ny;8}#2lk|`peH3}jLQOB`H1MycM++j%VcZXz zULFUA_VuMTP}wOQ)SJmva(G4y)-KVMYN|@J4K8;u z0zGe+K8|tquhjDpZhL)j^X$sU{bbW#Zn1b~gJ1X`gbPu^s=yX7ZvIWEmdwi)yYNG9 zj#1^oo#Uu6xAWE<*G~Q7{*An8NfZ%p2$D=j(&99kN6Ehp9~~5#3_J1oFI^fdQ-3CB z-Pyu^ach}J;adId7g9V##%~c+!;Y1ZBE@JwN7ID$0dL*@V1$A=gP({11y|&Ee;hWv zSwgoSBTVzz>tHC(B#pf$C0ELR*A4TxGKXQ)dKV{!Rfm)^kMy4ADf>AEzDesPXU)~> zggOOgy%20l!a+)9!|DNBgqTK>g8ki-kbz+v<9x$;m$6>|XBysI_3Z0)O5bbuk4t)J zCmk;b>iT>O6_RUMg>&XkQX5K#x7V}vNDqw7*^Fi6%#*6!9YfOOSdjj9(K$RtZ(`32 zIgq_@=IL953D9ZbJcap4yXVN>8fk?Cz8%mOh~n;2<$LiU5gyF-Z`*jrro2;estd5^-rw7}t?N8l2{w=2;0!c%{zu5}psWy4sZ3p{ z>YmUod}A0{uE!e_!Hczh8jvgW9vl|G+ew4fI;sx@aY9%baiwI1r2jxi3{p@}zG6=k ztvs{q5BSIgWxKxx^dBJq@e2?(-khq~XHsK$gH5bp2-|H*HFz+)%C~+_@R~46dlq2( zrBhnC`MN|8I~L0cET?=GTEI%DnMX&2><!A)j5V)RIU6Gu=dyY6+{Xe2*sE5 z&n336NAZ0KCJrZRZf@8YHYh~OP=h&-L&{1m+va=zW`E~AJ-^dZ1D`&1T&?R*hwoFI zfskXrJ?h4vrVJ(3i!T>NjJ>&YoMO)SpB%?Wzo*qXqO=d9#s~+i5l5~N?DjU@puMZnHE%7{T=L65H%`AN1*QCrW zx35O0A;}I3imAS&=^W2F|2f(m!~139pIRvLBPjdF`&4c=W*HuI*WxM%nku!@orBdn zI9PdaE^(FRLZqDpjv-3V!0%b8p~>&qvxRDl5R8VF)na`9^?X5ukoQR6V(@wl$Slgn zx2w{{5_6B$4gkjHf8#OWt0M<+@v{~wVoGE&Do7v=$#9QjDb?IDqY`z(FM#KTE!cW3 z%@JJ(sN52ZsOv+~>uc&}$W~D)M2G^ngA>vejn~R&Ke>P=zb*mDVLS-k8;+P~W3JK6 z<8nJ9)UMJ>#k-)!>0AZ|)w(N7T9URsFY4QD7|2I=Ti4YJk%}h75iMSf2!e`FLsHUm zd2nohyj4QyhhTHqXvRwhln{Rap$~E@uN%g|iS%1-^x{Tava!wK!<7vi2f#`L|KrXu zC{z7Vx~$HTyZkYA=0DCX5Jg3+2Fo4+`_@U;HLt5gA0`#^q2vEJAyVWw0LO)MUydm$ z1yYtY7=DupumTk*8!#^C^*5p`?)$FG-5Fu=)y6mwY zf)W0y)O@G;kQ8izU(J-y)(hrbawU8ImB`eh)^4edeylv>NyMhH6YY=qvA@rs{ z8GR`g$%QEe!s48`Z0})9q)n<+$6q{e!QFbLjPvflG=UVx+g1}a^>>9IcX;G)&c1p7 zeio_j+c5&7sW2$ct+6mbZ|`fX8e`1Z;SY?wy)M4O)GX_uHbxx&}T3FI=yca0RmqhBv_81?X&8==Q9-qOqmBH=G z`t6}xT+|Nx#8N?DL;%O%B2a}FV#sDJSZ~_Ei%m%#lusO}ZT!ds@dY8W2O^6jC1b@MLZWPQqI9aE9|>F^+$u zBiz{jr4Mmb$coB64HXsP@QOmHXe2jhwt}?QF zGQ?p{k5tioy}PR}jQR+Nm~Ake4s;xR4E6K*qD%Mads_E4LWfSC-62|HM)~#-L+Rw5 z;oRwA7MC9d+dA7e*Dw|pzcIZ4`fdN515^!ox@&p}ykC10+;w%MXCnzRNL8IizW;i* zJzr&_PZl68{=fz?H~_1AeUiVNHzbsq)UH)>uMA+whEIdyX_22^i%5T_LtnU5ocnHY57&&vMh#y&i8EJ;n}J~Vjjmd{7`Cn!PeFg1;^EgHWrN|3M{P&z>3~ zf9I_VNt2%d9)F7oLfgLn{`=pR6EB|+`qDaasdM$p44srG-8#*Ns;|U6sI@6{$)DC~ zN*<6h$ZA}*QJ{vse}Er}D{)aTra$tg%dE+T=D=q>Il%1q(2>j6Ou7S~2}2IDX|uYi zVdaNsgvCCqPVu1Yt4e>}hxumLyTC`Mu4~M~ksxJFX{A-Qu<<5+Ty`>HA^L<(KPW(w z2iaxHZ29nU@V@DAqmQys$k;357b0t(E6a+@H{>nTE5I7tU>)Gg&p(dzVuI0WDd!>l z%fGZcl)M5URIE-L5Qf~x<>?MQ{VR9bBff8M89ScvEpki>E8teNY52j_BKXh?EkLbm84%M-J@G2CvvU@DEVUh zeCa2W6XtemfuE&J(tXG5tuvb4X-``{UMUw}4-@Es$dEq*KLz$)NAsSI69(uW0-<19 zO=~TdK;L4Tvlkm_$mq);l~|9!^gP}SOo#=e3qWxr1T6l@lyY%M79{Ns5Bg)(Tcetk z{nE(|VcR8}?CyEKlz8&>h2~tM=BC4;719C`i1$feEY9MIMdp`@AZmAYhC=&g@M{@b zIu4H2e7Lq+HHog`;UzOM<6eW)#nV#UNr1}ylio)#r;j^Pr|-W6G&)s#k?>|uY>nEG z+I)$IuBb+XG$xIe6u?hZqfwmYa{=4sdhtkMm62cK70NT;TOeo87NfrU z7|M!Ym(v3zxW~*S3D0}U#)H=(_1BJKZaX}A{Wr_cmq%|Fdon>4n{-60Xtp;jWp?L7 zqZum0@7{a7q7t|keSPB0vHI2L%d6Nv3fue){q(~9y`2b5lGObE({6QsL^XzW*`Lc3 zqwGG$G_{Mt)qV8^F?yFl)4#KJ{fcF2^)op3kV_WPwme00Af?gP`ulYjP*pV8a1>zl z)8B2UNnVja%#sm4^~?iG>(fxQT%*#lQOFRGI>Dia{=Wy z8TdPXo~rXSAN}3z?5b-jUd{`JzdTn8cygjFJVw??e4J*0R^*$Eqj4PezqjR2#bff+ z$HT^gizow%qRNa)xKXUV5G-Lja)T=tq)BhY;$IsYg$_{-m8vy>H0y`81q#_lWYjAp zV(zIrLp=_sBj|=}9+sO<-shi*CksQvuVP?-gsd(21+XmL_yt?*^$g{mkPPQ)7co+@=vp z-SnIL73bt$>L((MSMG6vjedLMlUzgWW?*GtT7Bc5>CgfFy78p4K3y0LAj_SJj=pXO z`wf%4p0DDIo`=rB(nW(^WMd6*jeKbtT&MPst*EDBqb-RsfkUgKvSR$aB;1bFd9Hv` z2!AD^z#MFLFTXd?i7;#k%)NQ~tC3B}C1vkP3e@{|@@1b0D>W@fE<^LDA1%W{PvdVk`3J9(KN?oyXQ2Ks+=%<_9xdbt_R&XWC$iyhP(svV^65d(Ot~l{A3H`PL1-eW;)4{(!+yDvvPnZ`*JquXw{>5 zDMd0TtcKK_xAKbNnvnE`4=rMW@n9H}%#DON#E(U8PC%jv`$X`?>UD$pbOLlJjN8>ud%& zO$52PLy>#A%U@@$7l}xpI6_3e%C(Vpd;Gi9df>gXP>%i3eBUf_GKjkl&|SubRk8Z* zJ0*;5&Jg3Mu+9_=Y0GSoQqttqJJ2E_ns8FOPT6KE(bpgF%0oHE$bS-t5IctEyIMwt z^7r2E^hy2k;W*56-S})OhWAxsj)ts=Ue70anjpF4- zYwq4}tBVXhp6k2TydtFZ9XNS$zhqh&I@EhXHqyUAE0Pgex4}<*drV8Ab*JcEuWnvQ zx1!`3#t1y@ns`I0a;8atYN;;ajWrFm$C$IRA24C((6xQ((~$%8ULVHrXGgDGt$wuS ziN?lk>c16q8-}J<7HFtM-?m6w(zfjHBpL$1_ix)1vf3?QE7S5QQ0BJx2H*hx-q5V4 z#-}ZUYDBCcEfCoqP2t`Tok0Bulwh8To*neA>NTVh6Y7d1!P^`*S#3t#mrkv;h4Tlr zrihpl-afu6X;QX7BZ&sNykK;XuEI+g^3;Rv7-*E=4e7ziv{7!uMDS$fom^cplUJ9_ z=3Pq~%;0J)&p+Jff9jV{^4De30mjh;KL-)66y9D!`@pxy>#fi~S0u-7|%?GBdq=x>NRfpWii- zj%lr;=aZCZvb3{K$GNCDpayb+_b|DO=PL3yHx60p4KF#D` zLPcp;X11CFU_Wwh5KlzrDe^(QfWE-QS)#rj%O>aF4gcQ}b|2HBzJY!gwxFINO+{lx z-%KG4TIi-N=IA)RK>|MP@X~NUifXk1sq;F*=Eug-Z5zPsEOOEeG~$#NUiG6;@uHec zd+T-IE^9#TF^#6RTCt!33PNuby5?5w&zOG3p>9SbFU4;7-s5toND9-F^(Ys#f$rz& zPoghRCm&(cav4&VJo)uy5AOyw`480|_vVDYI4R^so$rciL`X+JcQo(sIlf`~@}ekz z(EUepgF``(_2aCO7?LbE^73XNHHj6C^ld+Kt--Pp(cev^EVevxvZ1M0oo_6b3a%j! zMh;}|RnDrHfF##+w&S#wVGm~flvAWDH)-3@&gInUh9Fk=wx!VcJU1|knX;nahUw=mpPiISTjQMm0u7F@ zxvQd4-T4=Xrh@%jpJm3zQLhYdt<1#@UgM2Yu|bu}xHR*C!edIG5Cl$_wNQzr)mK_} z$7kl_BaCt>Ux>Np4S;gg0j$RE{u5*K;oYYbIYS|M9)eHwV#ze4F@(<@54c8f;)j?J z+NoFgj<#EJkt7wOs=T(jMYS_!C^G%!yh!4Q*S{Etf0O*|aO2>#wQGf6nV@iBu2!(X z%3Wvovf^B6`F_W9`$q_6a`e^*z>Mc$gwn!93OK(1rJ!G*R=s~LulFMrsGc|d$;US#yS&=ysF_Of30`PF0GumLMzf;Al=D|bUyT7) z|MgxVLxOj;yVlng8F-j)GL8=@T-%sy;0f&&q>A~4? zkM+rfyahgW-v)kufXI9_-><4HS4NfP~If#ro# zLm_^<--1df4-xT#G8yV?3Rm-gB6iiICo*qKd@^R*XmxUBT2?3UJ$Th&5x4&jf)sMX ze*4Qt(@Hrghz zJpY4>sYRg*d|KumAfx7;BBQTHZ~jJ^*H4s_T(VUtc-0RYT#Z%t;4DOC+>oMRRPGL| z1gn_oB9YB&e6Zr>q>lL`A~G)Va?-x6T9~$GpxuN?kYhSgs6IYmc#ZnSIObpiy^==4 zvjxqeV1h-CnRW(-5s=hD9turVR?+r+j1}Q-z z_%yyqp6aV`?fyP3>3D9!#H8}cgP=k2=X2Q--|iVWR2AIrgQb! zTV-{Rk=3t$@Ac)iZGtk{Y2K5^E8>xxp?Khi7sT{tu2kSQhsAy;{H#`PN=A(8)jrs` zIp)5R{wb_d+8t;?`#a@Q${Xx*u^*}jhmfeyG7W{$Z`m^$6+!v`p#QjJd%oKIYe!I78W(GJJpjx57?q?-I5X{?%vZS?Of$h~}C51hsTC>mL> z#m{qf9KttS_SehLCNBCBT~eZ;sXs56aE6(yqSW!M zW9%Nk`^8WapR)58h`$Hxr=P1<33_DZQr_VyMJ1(i&>M>FkFoG4$3Z`<7F19d>k_zW z=DD9n?)yHT0zIL+V7vjDvt~&+q6^;3Z`+(sdG2p%waR=X$MWw@Ne7w)DEnjSC65j8 z#SLn#=rWbp0m?tWBqXr<#OHavI()KBE=)L{Qeoz0)2n})tAW;a1J4N3>C=n3eAS;z zd8jfzH-!N1yrRks4Yzo9#uT2qFoPdMRGhY6pjuE2iKb*nDd;nP1vDgk+;82bXu_ZC zmBS+}B5Z+J{-uKU$hfg|Hh6j#PWZ3LWx@$J4p#r9*J&(G^$PN#{RK+~!(=;DEFLNH zUsxkcGx~Q_sHMK4uRwZWr*<&eD0gJh=XEaoRO1E!c~>2>0EkIh+#s$bzT_g_SxN&B z{=uEoPPk)?1iDYp7XlN2&Y>w>U@Q~y$aq3u*M;-fipqw3*@L!(9 zEg_1V1lv*0;zfe_3}Gqiuf>$=HMquzZXLa3c6RuKTnxi3)ou69)YeyAJ&HS6{{qqp zq2=ikEzPNltZg+8@woh=Kb)D61GtXw2tzblUB_VgF8=>Z2Na7)#_EAW9)|kcxurWim_Vzb`sL zJsjB5Vc0*bk#pL-88J>36OpkKxLGGBmqplgxQymb@w*XS;E(=7pA%4#I$TKA z$IpA1aNFF)RY>*f^PG!}W*27_y?KZFZF<}#o$J8Q<)({`!e5?%mEhGt#*)KL96poC zMc<1J0b_Dh!#vv;&{>B-A_iFhFo z_tw?+?$X|r%7|smAH;sI@+0~pSN$0}6)9vvRq^nH$lhs|ecT+MjPn^hsqg&LcpLJH z%@>)Yp9RxUt!f%gmtTHvRP4*nQfZr1(Bf}>@BIPgwHl>RVbDO|+2zd*K7?}eoJApU zgzO{eM(1dW|KxqlEt4gUTW89Na_;sE**^K}CI|j<$`-J@oPNd`FLgwTTABV2soO{f ze!xIO@H`b9($`rIi~w{x{*>=UAE~DzWW@2_Rvzx}4xx#xFhvLEm>NNpH(R`>b2nrd zP+vSvESCcVR=F>@H0zYz-;1ng1LX2#J>41sJ*EQ8#f$N7?)&G(uvx!#e`oVFe!N+`-`k zeP|t$1p8g$^dBy5Ua{y-*Lq$zK2;|w)xS5qZ35z<4ww`3=SlPHi>6D8JlBJ6+q6{u zF@hOn7p0>_a-bW%><>CzWZ(y0YKBu6n6(6&^GwVKH6+i~S9b5)=={xZ+|}hDyaz&D zDn(#39fYOJDPyQ`xZz`1=l+s_QcVA`Z79?ax-L-+lehwkfwUf2P3Y`a8#$b1H1>Ha z(%$^bK4d?wPJoiTJ(O6g8c@#cshNG+7PH`Hq;UCCq5IsGtZ*JT5qTJUcS#!T=cwPZGh5Q`gsU1%<$L_9sU203R z{~kP6a6B%(pX2QySM`8Ga=?74HjN{7C?Gq2O!H%&G~=94L=Qf9tR(U!z!U(+k~b8x z0)j17Z&Y8Iy3-Y)1$>=?}Ohs-Ut%si2%G*&%W72Q0o5Qw< z7ImuPcS71|DZic62z6{m7+z*K&! zg=tP(Hj76w!F_C)}zNAM_tvN z4nrQ+txE{QcMKafgjYhjWEfb+mpm9VQ>sv2H1}%aM4hOAn)6?rwxx~OUFR=iAY<+I ze)VLT^um6kvUVqs^CZ589b#|J0HOi-GxR3KrgL=3hpqMceY_(C=u(B8LJIJ#TVb5pIDYQ!^Ha{pp{sX;0GuCX)oa~M>*(acj0 zte7nFy-!WPEW5z}gj?hvs`3v|LqdauZVr90r&Y$+e;sIA(&WO#su79GN{HNmsRVo5 zwILzp8(xMZm75;saC&BL~2hO?2nPwEX z!Xmo6(S>j)*AOg9blAA($e585kaij5pl|g4^~2BX8{9X^*cr87k>7R~sX zB{mz$l~x_E!h(bZ>Im}dKW2CrL)**YX!4p_^7kv?n?4fDU7m+4cUz-P$}#T>{k^`N ztu!I1V1U7pOtIxo`Q`88Fo?&^F~2DFuuzDz*=D_JGOkK^j%0)nE9mT(uQqg`Gej`h zF?I&rrfY8{9w$lQFH%pL2J@<5Q0FX%blOPz+z(f;)`_XiDUbsAK(;?YKT0`R9A(in zsl7imGA@O@|CTe-Gxv z%ILO+yk$^HJj=!aTOqyfe!QxMPM|^ISz9C|o5anwH>Vl;Q#wa0HzZ()({`oa8lP-U znLR2FF;k{fL8E$cCER^{$rrU&2~c`p=G@DQ2I5l!{{w>%DfMBCD)l;*{&=6-xYeVV zKO=16=_lk!Z-s?6uDa@DeO9G4{bD+DwJ|WjAHlE>j_DQ8x9&qN2YXr)!h1z$K5dyD zZ-9g#&)%@^7kpw>Wh?MqlHL5?yhUs?XHI?+`{OZrG5ysHkbxs@$@H>0qp^ zb!&9TyKa|HyTk@kX3fNRJ)3;xWuO%N{Q28fP(gaLWJdMU zg;`;_e%2?jLg+nv8t3+Aq<2X8_2BF|qxDAlSSqs;_{@d%#8c<1H^_)ek{eT42m{Fm z{8k*dpr|SV%UoWiZ}mpVn(En-;I=RZ zk_C8rT4M|K6l)q1NaCK_HyINvVvSG=(Y|TnnRo#-c-{>a(`ZPQF;0Af%y9w`N3+4zQ>O1LZeiYEnGooIGjkY zn_m``WHiDr8&E3JynG|s{vO3l=5hW*z6*Og)5rQLksk9E?4BG0w5kZFayicz4!xMl zzyq*SLnO+xp>aQ93Ro}29yFAB$(PN{T#QaW+s(GB8ZB>rN3Aoi)U#Ue-wL1L@P84q z%@)mQe|8hq3=bCkJ{YJ>^_dCi_(ZJE6$7rbh2!ElGA1MAQ zXe|a7g8>C390Z-W#+eXRP$XXdK(pODQ(&!SA^GT#S8IWLb<y6Y;3|7QO}m-&#gqgMjT}JUR5FzXZ}KoM2Q7_Iv7FNAz`+E`tRf8eJ5H)a912*%EQM~&E8dU4YwdL5V*Zm__^8?K82 zI-r1Z2DJB?^~)(aQgR@WUU~#s=tZV}T%<3(ABdxsIo$r$NF+3MjdLeDr1h1W>tp%``rs=wukw?=8*4Dp3v^6UcQv^wF;|3H)Z<#oVpnVYTNw80siG zMoU3VU)B8cln^zJs9r|(>9k;gDOK}2sbuyoS*|roMaDv$KGISC*;R00()SI1*4sbP z*}GZvMCK<7N#+^eJ%z?^NOngLw}O@(Bx|gd2R?IN`3@M-!t*xKROB? z4Ja5;-1GX5pU$Wap$<~2V)F)H%XA(0Y4pmp`wK;qL(PZZ-X|#&SJG)!ip^XqbC8K& zV;W+p&bsUP%!Ao8Ob%)d;g?e+OuU`gtsLE(`Fmd4@RW%UIN;`$B2G}JTsQFgNgH)xiJm7iex-R9PzXm(_LVYUI&Gb2YA<)2$c|ImhLChXhJ zILUdy`Hs>+hAq=2%Iq0sy#y*GFai>X2*!18 zcfKt6ri$q9LmQeAM7|6Uej9R@pa*52ojZN-L+ejRLX@-XzeQ=Rp&v^%=i8>)NvH9I zC(m#xX}q`oOd7-)e7^eY3c}>n7*Z2(4VIaOhI9$`@)H9o^=NL^ z(M?}#E4PW#w_620aF>w;^rc@p94Ysa*}}1SQrw&fe2Ct3%ob=bA`~BQ)yBBmXWI3! z{iWLkFSiIImq`}gx|w!)!^G{|DzEVWF?E#zQLsywWFy3ey1V%n-}m18{qzs+?(@vdIcLty1D*(hy_opp6yQTt-I-^CH(K)1-t&+n zN5L^Hci}dX&Ep&PS_c8byD0+3-z-p13Nxp!`Ws69ecL_6kdu7XXKt|$Tgq@iQz&iM zFVJ~zrVf5&kGhBs5}4pG4hg>GZy@PRaKkw21CY}?`D=~Ea)a>A`Ia*`EdXjw(D%0O zNU%pv;gvALJL+-?q5)Ca{T7&c{i|G{-3;%7x<-q~exOt&T(m0&nM{#)ybrOj?E@Os z$kHz)`N~2@V~%|)$_AP}jPc&5(k@-uNC!tqPJ|5JRAANZW1(Ql#}BKFoCx4rnuQKr zE5T&+e60976OW8fa;yC?h4%Lob|XM7*BSrtSKd~!p}v24csEE!cJ#@C79574NQfe< zstaU0C3}9t82G#IV_o;`=>gA(+~bf`mA6O%+Xxo&H)I8IQD>FV@j(7>H{Js}eu4ue z*$yO{3u7aVKgU>3M;&FJGg-Ls=+mZAKl88m)cq!5i3wMXa`ce6LW(*%im%9OwN1M@ z48%fs@T=HPo4@X-k=+1{Kb$Bb;3}?Pd!#G|&=QDO)^`?3h_OmE!2v4-i-#KYynhIV znsQ1B8Tv^*9o307#OsLoDvP(XsAQ;4cfQjrD61zHt|THqTU{3vch_%7`q2Q&R5@aI zVEA9v$_ej|F&(x0_TaRN!RB>G%IYH*LkdP&sOTGlS%~DI;if!Jfbwor_OVu4t)RhD zzdPBZIb+6Ni`zY9ziXZHxZRQmC(TZ?PW60k&ol+ArH_!Dfp`eRV2T|HVNvy=gbSt| z@_>*7i0sckTjNc!k6DieM%uL5#po`?%SbRdv<3Ca>=(fZyD1XW+7*nfSbA|>4M95+{zM<_2}9GIlWcv$A`mD zI5Y`$$>iZmO5M+oOpu<1$H>2t?L8g7H_jkH8Tf*6y@d;XwjzSfd3x97s6}`CapaE; z3Gq%QFV_C!c8#*swu)DJJ&31({m8Qqi5<$zqxGHim|U*UGLZ%{MW^%{ji|T}@6Yg)@G2eM8tZm@#dGy&sn24q3ctN zy0~*Y>Uh!n6!u~C5Ke1__VPkSP(!Exp4M#@n=p{$^N7&?ALbLJA+P32{PzBQ#20d4 zp-7&FNox*AA7F>bx|TJzkQ-U=?A&Ud{Z%}_a|E~@LYlBUv1NAn9B%W0X_vb7B@lwA z`Fkw0<0-D8#BY+4yd)hhjoTrtXK569(9rhq;WyVxAKa|);2LP$BFx=qTWJSBM2aBh zE$xDjfbZl`^WM+c@K*C23?17UJi!^vIg(6>f8+QFeL=4R|LaFB)2h!LhB4AbcEvO7 zv3_q^5ev%bHK&}ru7MVNP$ay2`|L}bgrF=733U$)!MlMZ{TVv_QG?sUI~hcWk7YTq zp(18c6p`{Ln~W8(plIRlm`ers&{$rd>HFAWQ);OP#%*Jx#weaHE_?=#5Ac};Usr!* zqWVHH)?!AGk?SVi&w1I)?ti>0kd>3yP~!f&D|u|>(E9t6pWcMyd174Vm44^0pq|ux z(p&+h zv!)tt!j3Shz8}Pz=s<=;oYJcjL3vZa=eKNU`}~PCi!*`C%>}2qEOQ%DTSm7Oh!U$bhqCRpnvjoPYm{eS_zLu$f@cx?#-4WUwZjaL4cY(y)Oa z%qM;XcIR2mqoqt)-!W9XFvP|e2&lh7b~G45lIk`$`;|qxq8FFgLT9V(WQib#H?oEH z1@P1f-z;C=#QqMtJqgPm6@=XX$q&^ix`V!B#JU8Qv1Mv*L5vW5mn1aM0BkV7;}j1- z0m^>#3gzzai<5^F5<4x}d<^G$)vFMfYgFPG+rp>gw{^%XtWNz@-$kC`9=9xIzG&3P&71rzO^!>nmp z&MNnywbvCY=Lx#6pg|k~b00lb*LERqV_pFZrh8aI1RDaRm6u_EW2roYHTce0PD-W1UnW+!|Ejbg$tOdn+ z#geYyBQ!e7qI)LkHJ#FVZXa@w`@=vl(w>ubb7rHTx7BBUJU-VMlpk=OTtCjg4zQ3$ zaR=yx>UKMaiqsD(aVO~qG06vz^9H?&PL_57T_OoI-|yjU`t&i$&;Qx$Bp6OI^1*(v zBM`yOLh=RDFj5;a6*cm>3a`?Tit2 zJ2=f@iKJl0aKpQTIeqCGLDKq?3ULbZ6S~QmC(QI-+xKitzsLbI?>r1e8OgCz#ABSJ z&7ZNmy=94?2D8*U#60qfc^v^>pWrelz1 zfC{^129UqR9=(DDjRR0!!#oRL$;3%nLsdbMNX(CfN&$X)wb^O8jg2nLKRI~~;-cvG zYuDpoa@3P@+kbWeEZ9I+n>SfK^QU3o9k6GA9PN+6oyK{&!8t2^KSnk8VHU}?ze+u7 zY)gi;e2Tj6D<%@VYNR~uT!nD#ZWd}FW$ywZ*An|TTLx_3Sqg_DQt{zAF#b_x5Fu6u z9Pz)y`e6K_LjZ5>UQk%%4U_!?&3;0Ig5EP3YH{Z#vrE`d^~(yQI+FPbLXdTPsBmP< zEeg^uGc^kcap2A~!8w({#EUJP^E3E59mj}<>yOvo_Z;n6@yz;EQcUe6mcIVUm%80yD9O4U`w#l9a|2)mGA6$N<9oWMcvN$tG8N_VmfX>^R z&WZBX*5yuXmI<8rw5`1(!5TX~jXD})Lf4h23eqwAa~JP3ice7?=mS9}1tX_h0ND2) zDt(Q&#(jNpb{raf#UBt2Cqp~O=4$xBD!;ivy(HY&BKGP3m7GxU+gPSWIsw<4PBp)O z372@>Lb;OThT48FA0&raz-ixMKIWN&bZk^o=!3p(qfLT#d!@|&syKfHsU?Szc5JP` zW#4ZkYb~|Eo#ndHx86Y5&g#u@HfiX4j6Js9A=O>&E=b6f0KuqG1pbhdWOd4_8HN;5x7#iY=iUl8GcD19)nOHs|R{wP%1VaK=@FciR=6 zuE+(mcN8TU!8OQ(LAOT_5x!7!u>A6u?|@Yp1@Pmh;~kImYICyGx;rB@E(h^%2$p05 zy;%%}S^(qSOp^~Z&$t`MGJb&Xy$z(9hocM0gC#agK`iuS#p9gObtNKFxhVhlIoT)G zo@S@iEzf2oV;VDY=CyIIo*&Gc*Gug`QTCkxs-1X)Z$Er=jr{ZsrUy%; z*8WdecZh|D#N(drT(t5W@4RXpFJz2uJ;VW4%~z&K`MCv&h&SNJ0+l9j{^dBmbAN)e z|DAcF84A)>w#R61+d)iBfrZH6iMtkWyJAw@-J37(u(_qDs#b zki|5FP5&Ws_^?PEFPxD3Kq#riN_A77i-)B%C~rACf6&t%mX~MA^+)^C zEN^J7^Eb_$>!3cTeTVaS2~E+__d*DsMA)zu^%jgrax$6e(K>TsY_>X|=wq?(IaKss z3$_!czMC5!%MMvr8-H{YAhD8i&iMbi#t9wxeBKY7)+4q|YSm!dPUm27de1=rjt3^W z+CR^>OS3|BqgFY(Oo^70*{?B>vJTj3Bu6pfR_mOGyz0xkSbo@{(G$KvWU<~Eqe~=t zPuN`y7yo-r{?g%7l0qdt2kwHsSl3Tp3y^;=?vBg?ONBf;w?YZZS;L#~bq5F1vYI!v zy}x)E+v*h>q*p`Z(O~V+{f8Vqh56FixIG;$Dqig={*ZXPx>z7gHBgrM>AVfd_5Hw6 zo605R{(cQPWkIpD$nM~I@PhwkAzKZ+8I$`Npjnmzd;SzM$rgII5}I5Msx2K2=S)UL zPIyfRWY)ow$>5!RBoUOpuJ;MEx+AF*jKTKLqu-Qn0Mz7X{T}=5DRZkLG_BodD~0?& z!|iCt5S>2 zTpKTVF?J&tOZC7p@aY8jzLd1q2@N%|o8Ih%(s z997RC6TPUmS+39RAAa>WBLM_pB0HWr7$b?wwA>yTb)ubn0U2_LeuBj$b{kM!kJ0vS z7jV8Q5jrtIW9t*a-OZ8Ec)FTvqW6O&Iv%{$C&ewYJKOOm8*FuJY4|7#hJn;_H8%qx zj_;SL`N?@;85r^d)=%f^tDYYL3Qd5su`s~|&lf$17@-Gf1Qc(r<%IOHz$w!2Vbfd> z2+@2AVG#u>Zvhj~9rt)!!!C-*+fo*FHPU8{7pqqCTUBpL9eh?l;Xzsi9Qvo1_;2@=<5 zm3vL2EY_RfCwTSe-g=3|d~p1lNJyGHgv+Rb4#_kuX}KAGb8|>L2wGWEI9Lt+;+u zFz~D?PdnvSix;WOZ|oW;Ge;4hmLLbve#1~%p;%c^TXUc+KipqGCz&&8_3LguS~U`= z5W?D>rOU2SDEG#jrV3-h6D`g!F5&9kPO3R}UWrx2gF#0(0f3pDm#t@6+SKsZ$V^jJ zhg9WXEOLhm;^A>Mbe+t7O|W@i!1Yu64g34ygO;64TlIOEZQ}YqJVmEFt|Vw%_+D&p z_tO62iVhrpL9>mvy)(+}WAR>s`;=-tlRBJ7U=GQkO&%$4>iI8Y)&i`M2%9>*6E5@; zPxF8%s9mt~bs(Re%yOk2V*1fFrue|~r2KGVoBJ~>k2tJkMu4h_LA+K!s1MQUPo#5M z6TNu}r1hBI9K*AoquDaY5u;F|mA^3o8{9s(D~Gvo*TAQg)xne~0aVpkMq1_?T7|M| zm9UBVDl+O^?roZZzC%hP(0d2io7AeQb+ZsAkdfNR<@UYWuIg)O|@l;>?Z+88)y# zkX9Q7fJH=dTH6Y%H@j~|SD=WyZ2YnEn^)9mKDPrOK7%VZ2rC+gx-k!Dt!mj(ZaHbu zZ+8LC>4A;J+0YkNsZ(9*qqdLve5Ba==SC{ClJ`WbOwQo@lP$JkJe~>UG!Cw0l>?h7 zh4xS0K!YH7O{$&QJmV3gQtubaNO>v=5mEpQGSoX)4CN!3@xtNqN~Rq+%c;(&TW=QZ zYJDfq15p?ve@l->DxY-|(GP|KsHVm{`k89)Lbu}h82{CImjTSqUahDZ>n?}VB=(+; zJX@%d6{0fJNURYXyVBNmA?{v+eS)Ty4iw@{%9cRZIW7zrh_qG2fs8pt`q%l@H;|7w zRh1_+r8pF2z3y!_5$JLr6w#d~z^%JQ4@*PYOCZ>{9l0D}$-sl=oLn%Ox>RLvLCx@5 z+rb!JNg&TLp0YdNNyuC&NizCn=bzrz-#~g1(>Hs0YHsrB`ns^v3z1N(%lA(@B<~g? zF$&(>8aC-JSRPXw4lg06ovNR8+-X!DlQ5~X+8-TpZLTt)Qj@c|9nCyD?vLC0LpumPwD#ISjoEsK+(mhUQ zLMA2onwa>3{6MY+1Kg(ABeOF((I))5Ffk*$-@l!LOFz>kBs+`~@RlikeX+NcX{FNe z=!Kx(yGK)jPZA9P(9U@#X4{K=gkSI)tHblU#-H%R)qy6h3p^)G)i6!5;ZQf;=ie}) z?yoB`m+##sxA?8ok7>9VM#RH0^pv(rjYIq0l}UZgCZ!1cfAr!)QCDHG4RLm_T^`UH zOF}*gij@WP9z2jI1rJqswJ@S`K@;s+%{?kRrg$$%D=EZHHgkl(!7O0b;B!!~x~Kqa zmDqgHhi|Dx$x@^ssP24&6qi777aEht^x;j^GwL;Qn8!-PlUDboc`qa|7AGgXu&cG zh*6bOV*PR#Rk{)Ry)cwyfAwQ?0^pv!{*2Vd81yc{F7@ubN*u4bDIG=%Rb{;?n?V%S z!$4ppQYAgra}@m=SRNs3c6B^t`3N+{LNqbRyCHE_Xfeu<>uWEbBlO&LMTBPAT^ie;BHC)usz;Lxrm1_?B&U0kP~ ze>)>0n3I#`ARSnm-{T|!A(NB_IiKMwu>Ph;h!5!G+QtVnR5ViBo#DXxag4WSUALJ2 zC4%9v0{z$&E$!vUMb#d&4Cj!e3esa(t~4=%cKpZGUb?t-r;Ujw&FDNd7M=lLZn2@s??mXiPF)UuW5e(4>lq+6lz0x#u=Ofcyww72^&BvCg#W-T z0kn7bDJlE{nL1^6^y63h^@8{hTqDl8|G!Nqv@!VnxR4tx!q{$xib03+@)I6Uql6^8 zAml_85~jW1S=4R3e(MwS5!1#GqiZ5maRv7V^XNccGSsXX`uc2aaXF2ntNPTeTgTE} zbScd~&~3dVnT{TIQOK5)edQ1&Yb&by=f~fpk<$~`#WKg1Q0KIE12M*j@acE+>-9#k zhQ5}@U!Q~dh3(r?u5VI5ZBH~l@MY`6c~rbLN5C5k7s~ZrhS;rel-R!#W_w%7B=(SnX$-}hoxAkrO8**MBc=c6NrNO{$}VBNoU1QV)%fQzEPh^j&EJQyk2 zTDv~B%{W~8smswvG?0N=?ud|gkZ08@kWhXP8bXhzhuUQQ~y8BIpE61OaN7haM|4th#-HXN{=)ye8Q#=obs0YReZPI1Lh&E-9quiV zbAP$oCHOtk(ZUX+2CQKL-hEajw(*=Pbvd+)EE%7q$o9WXBcN%{Gh~dlIBHU;L>t%b4 zEuu3SAp3>~G=*z|jTiE;`;{&^k5X#4mhAl2IzfpU@E;@rh>+T)WopD-h0_sKiOs^V zRlFgWbTavC{{uRr))w(~^PxzHxn(?L%wN!z$PRD+)O}5RR~@q;wXh$Ok+pz8*Y{mL zgQK54$f(eA2GC*}r=|cHitsQtF1R%vg6j`^tVJo1ASbs6B8t!7Jn9&kZ+-C{+upPb zU7^A$5X5k?d^{n(4u`VuJV*&u{T~YP4C(%p!e*aup*)RNK+K;Xl*W9P%3Pt1<>!kJ z$x+)W?NMBVGd_RNmRp|=XCTy8T+5@@PJ|~NXeC$WBJERHco^uP*y%$wkfd9LviQ@* zaGvg^VC&DQ8>U+ZS*>_EL@UF@l1l_c;JhPvY^|aw($<3?0?{obBxJEX^;uBHP_@rH z^^D;snZ4&@%5Vetrvi&cDKF3V4_i^rO;lBy*x!;EXWuybo81B#WV1iDxhZ;~&=d!C zI8Q$4IKF7H{U%}$ z{4SB-lMN4#{URV+(Wu9sHCYZ8sJS5JP&l`DktJy`OIClqyC-yaO&dA=x=5~2MZ?@x z(LuFS<+D`icOyONFGx5Np`HIyv8J zw!K<+hNBS%F^&`u-*VDQj_AOK7|_z%1zI23D$O)Y0fUdKULnWL)Mi_&jM~1$4bGL^ zKt;%|JYx?Lf&1emb~ck_gykyIoOq@`j5%cfp_E?$MvllUU(sw$1Jv%>Yo-kd;hFBc+2%$m2Qf1JfL$bXigb>=*KKNycKmufyW5)`w@sbChADFSfkzvsGJ z=wUEAM0=uMpeR%&5_R!=!`^MHI6z(IzF+XhcYIdNmf|O=gS0YzXX=3ppH*xuPwb$Z zXKrshFcTpt{Z5?zKo-DE`DLBLog1R%o z`x!lkaaLO$M^gj7L=gPjfO;jdiysq*W1MBpR4){-OOxV8G~ZyWX&T(5wW4baC28CK z$V>RiNcwzqdY5gUV-mQGOaTA_Xw`gvg}*p^4kb`={2m1(57^JSU_c8g`g5tZ=`D3~-0N2}S*k!rjX5*2$;!#T$ezzANMROWSj#Nr;Y|7j@=+BJ8 zk|o2s`<6$k_f9LgMYpMHM+aW3$W=uO1niuz7At}^Leg)bMDK)Neoa0&B{a5ls3G7= zM%s}5FZ4(N3ur*_!c;k@V@rB|FImkTsEaC%ytj@Izzl{a5T@${il6!sf^Dyrtg=?P z?BVwnOotDon4$t)R7LD|uW|4A%_M??(lx;zcjS_{0hZnGQujjTfv8K_(YcDR8^R21 z3K}f*j`8f`S{}j+=OMQfbc2F&EsjTheJPNDUYG*?JV8K>bD&(H*!kiI+SZ_ zBNm?c^M(W&Z~t7}th&BcCNk(_OG(_$BWviDo zzG>$h&yck^e3nay_>h+|V1KPFW1P5qX>Vh!kM1H@%^~=S9qr>6x3#-AJD)FyJgp2Q zbCuXT4@y|&ck=H^gaEwVX=>E}B?W?LXILO*uZNAJ_j-F@^ z&R#bVmMg>0d*A>1Br}-CiOwU55yy{en5djS7u9A^MefY}b`VUI6&&?{U#y z1U0nLJyO6`z3hDJQ^C{CyM~<<_qd1yGNP0GL#q!xg>y$@O6iQWtdKOMXI@?cn*F zp8~tOO+v~;h=dpIzZSh3ykHg1;j>Nd46L^kYXhug*JWH?$#~x8TzY?d?JXJbP=f_Z z@M8Vp-#oh^)So3KOQhNQ01{{3toconVG&uARs-Z;1d%5F4P0C_JFFbn7sq+pp2f>g z>|ED+D!^O2D3(~Q7-Ledu-;E&S6U4>+_&bEPL~_j-<@SE#Csy zoNC!_KNM_g_c!!$zaWV<%;|Y)W|%AJO&%K7);~djz0%i6g-sBtiK_Dfv5CsMPhKJ& zWc=@kV(^RO;Wthb;OYlAPuc$~A(O121J&!ODj9Uk{Yj^Th{-N@>(9~AVSXY(awN-Z z!#lkUHU#wN1dZ$%E1cYx-jwez>FlTyCX67}uPc48CBLt$yZLxnEL%K>tO@6NR`$&~ zE6G9hPyx#?=-A4M37|9$r}x4z|!u_$8+UwTMgY_%Q|9niF=JGLbKeGdq$0 zCmP6T8^+rn3XcJqEK(j$FrNl|l=A4n)h4Dfw4S&QijQ9a;E|GA@B=(=FcSe1<>58+ z#>RF(WG=p=pVAOk&g52zJzLP-1XB@?6pCq5V?ob}>gG%DIW1saDCNy~aR&?e@*{j% zmTxBUgW%xD1MY;x=@zexv{JaUB%>gAr+&>ok`D&@S1BzV0Ex_r^Ny3bwNCjlcs`=_ z1a?a1BK9*+m;U2iWsg!dru65jqs4^3opuFMZk@+x-5Zy$+@e|hW8N6)Wp3tttUF!$ zRoAcAtMG#0SG=Tq$Bg2Gu#mQ_3acupL8&6f-ExL>k{P&nxrq;(_rF%ULtoJ^2|T7@ ze1Q)I=WMa+uNB}GToZ4>5&__QCjEdQ5B|&K0XYm55eU5TDY7W6(r#pvDL(n~k zahUJ8?!V^|b|7s($#h&SR``lR;(lp-!Buei{%tIME@QgQL)!fn3aN0sLh6oo>?EcEM;>>dIoP|2Oc!@*#)!YAGdHr;o!{JEdY>N zNBWe9Wi4j$^6^aAF>kJw*MMQN9J8ZEYQexyLmUdca*_?A2%;_tdizT>bw)=Y&-`}P z*(WLcQG>%5421*V806r>8b9Z>y_s|(Ufd%0>0IPk=PUDkpIst<4(71%tt8{Kqc(t> z#F&*T#y!8n6JD9->%Py52x{i^%vQ!Rs^px>yT}NzlZ!HYVx>s>xlSG*_R(PGzu!uZJ%Jn--sN@ct>kggG-(>%W zDQ(HCpHs}vHm)pmnp#Z9L`ta)Hn=LZ8nAw%L%O(u^=%ts=fttvj6vm~BBlCWcRZTp z9~xG>*TPBAr%d%WqG6 z(Q;KXH2OUg5YDG3F1}U>aB?YT3+2?v0%jNE{<0%#)+ZFDw>=WuN3Sb66nPswnFH{) zI>5%1D5v3L@wsE2*O_td{{@$u5#2F340hjA7bn2yv-#W-J?rrv6sHFI`YxD)bn07I z)pTZr#XU)-})v91REkI$*pCR(d0ky&oJ<;gO^iihpT6e9EaN1$kKv9_YNtNNJ@c0j828*h)*#6$) zZNzs>r^kuM)PU$?i%-v}`ISb6!79P3q>(X#sqZje?p7>V)nitFLK#NI{;L_e8a zcw_%#0z?5FcD+VI&z3D))JxlYvI>zQ3U%AS00(@o=os2i;OT*Jf(TEp9P%Ym{7@nT8Sso0xZjEj3OW1~1lSr{^9dmnJj87{V0{Isy~?qb;%PDG4Zs@WAVeRj z_wDZw1q=$AHhJ4dfg3B9_s*m3y88acd6M1SxdBbF7`>{WnvkQ1!{9dR^gQ1%UUFb~ z938flHkIx(L}G@SFC1nCIq#NYJsp86^yubSI2No9AsJD+^8|y$jPLG@UkDl(YwRD= z1T)B%!rh;ckRFG5+agwJ4LQm*h%r-tZO5sZ@t#Q-l=$3=>VE;d?;pNTm zo%Ud#i-(G;?U361ZIrylr4Q-Z#*v2gc>Lfe?M*(9=xRa8I9)71!D{kul#iJ6aXr&Y zqxZjNugkxo7DrFd&P6Vtvst#`bw?rgsV--newl${P5wr+LWv@J?x1OvM|h7r5VKS~ zn^c58%nT1GxsiTItV~G0c6CIJKp$%0@%=vYlDXD|b)?|z?|0Zvt19Hl+h@Ry25^Pv zwl!q%3P6Q=)b&K?RI8d}04^u9NX)m?{X-7WU`~jDz8Uuy``}S)_=1x)(eh^dYj@hW zsf1yGHBUjt;#dC7r^K2F5Xzs7rZ}7&|L(as!IGT%5B8pj8JsGrZ?t4hxO+&!$p;s_ zgs)PfpvqSzXof-l3c5D`%RFqVYQn+S4_pMZOgoP$wTj?pX(!8azc)o>(xDf8u!fxZ zXgn=)dd*05aztl*^6^`P=T*8%JqyfN@G2ls z1JV%l$4Xv+NVvV%rph3HO?}+qalt=|4;Y7`9RXI>YGFO3k~kRhdS}amoot|G)|9v9p97GLCZ4ruJQ6gvR*Y_ffe zcg2&@2ZSG2puI-j(*l*}&iTJF)^xjTNlmV$E>>^#x>W4?L`F}q&^Rt*pF$)!q+MrP zCK5xGdRBC%DYy8D!wbEoh>Z_W;^5ba3PK(vO6!TZNgfw!PMj$!-g4kJx|`=ERU?(O zsis*?95s5BQrKS~n+Muo(b3x5f)^7YM~@)ZFJGO(M`kZsWgpC(C|EFS8p0-Y?x zHaWAxSyUfZWmjDrk(Ly`)72%5WO!v67ts&e#qzl=8vyD|hV@x2yPb5N2*}@;z?YsH znso;p=e=^~obUnbpE6PNXG(26nDe*3pia9vnm>GiX z7r5nlM(Mk#y=&{*iDdeF{|;)R18fiVbotW`4bOn0J+Y93I_X(%ETrWSPG_ld=<{5l zwY@Fuk!Sl_C>WEmbhoGkxgTpjFU|y=^wnw}cFGEUE`W7PmyRGBJova2X0CO=9O^gHnO0mBsxRhaU`+PbcniAM4fdqq!6tad?`%oh3{x8wBLf390thl z{DXt@Q%bTczRTm+IjyGvtWd23M|J+Qg23YX8k8VN3=#1N(#C^_${TMq#*nQKA7mTV@D( zb!qgqx{+I6eDq5c(fN~Twq^L8mYPBZxX=%$h!m(s{TTzwSc3w{U?WgO4BK(hEGSCP zeopG(KPU+_A*c^_05uobT_izwOt|^^bKx?J# zwL9pGj@m=8@t1s`wt1c4-O1Wo{KA`RT ziCqCE6UAu2r9xyzS+cxXp>3`~h&MSW7j3$F#=glXWt?wnIOgy*z`&ijESL;ul8Xrn!YTsG0J z2Q9`}Oygx2@`3lgKM#VC)Y*8l{=EQor&=!XtM+$@B%aKUP)dMb_eZEZ* z*POrO_HlTkh<(<0t1QXP8MJkf#!gCg4+2l7oA8}N&W)S?GmEpWXS02wK=db2OXql% zNQ_hvg;>b`dKS-B_v?Y{-2Nw2FOd-yi|{&>sEa&m!IN=99?7NXjR-}N57}oVX=1PV z)2DUb97=x`#5ZNtGpIWvKKdEwL~cS z@Y;|CPom>TE(IFsa;~yE0D&mts}}U{~16VJrX^7`c(oCrE%9WA|Uuobjlz`DQBjbLHPYfuUhQ)2NvmH z)Pb}K?JToT(a#86CfAp-%88aTX04B?AonX*i(k6zt&m_l*qA%QsJr;sR3EmyHDbND z?sR8Yw37Zu&PIrvo1AfT_d4DGddmfiN$r{qgtb$6)SAkH@wqwRV*RGXfb>86Di=0o zYU;Zp-KHj`9ECF!&+726Mo?Z@3#n6t+^DM80_Yr~Z8 z#LR~6<#D(Ei8twa5K&AsGCDtVZ9T-dRnY8^!3!wH4S8685ES(O84ix4>yEJPla$tY zyQ`7ild2VV7DHI8Ks=AzJ~q7x)VepLqB>c3$ua%}8>c`&ckE@>BcuAS$!4YZ3jkgd z#1ruGf}R-z&W|D|44nV4%!MPb8ABr&C7bLc!UR6UUfnHfZw@5?8lktQsgIS02F-+2 zSVC)!_wjLDEDi%gCyp=-s?J}pJyVs-_Pej!BKtepzLRkbRv$#ByWQ%f7ttW)ZR<9p z(Spb7N}p9b#l@yxpYtP^>Sg`*Us7ViB5py9b&XI}hqu1+MI$9d+U$Njn&d^|<)Ri8 zV~Y7zQl(1YPl<_$#J>qWZ3ULwRh$cah{=Pvd-ulcRM>&GMrrZ|Od5 zY<{8md!l8}oWrVO_KJiS?SZoky;O{Z&410&AO&%Kp9H$*8tqfq^c)XYV?6FpGf9kx z?-XHxM8M^ELdlYk^-bfkATs?<0)X9?YWUPDcl|P?I4X2b%4u}U@n>H zzhk-JN&Z$_y>vXm$xfHpe+;jqa^!DPA6P-bRQmMzC)*(O=k4>mno9~P&kCLgt({4@zvF5IsIQeiZj!ocAONgGtv!|EzIdm*A#i$ zqq*XFjZ~!KVQy$z%#uTYAuU^|vrD;W8Kc?DrnAYj*_C$#j#Oq{f<)gRiXklyhWo>#r+fEP?QsNcFwKY5~uG`GN1Q-_$C}>?ZjtXD7qBx$IAp>01jFCBTx3%0AiG|Ecza-{Wy=_fc z@R8!ptGrCO;W&=ztAOo%nm+Q~8cR-B?Qk<4jR|IHV$o-tX0M~mXg+9f>cz@;vZ*$5 zv7XH=_huP27`X>Pz)PHI1G7mbJ=IPomb6=BEjUb}pFU^~Vv~Kfn_r9TWa{2ui1yA-@H0{nDQVw__65a*6}xLC2X_$(6XjccGEY02NQk()B8 zP8*nZ9ighQk}0sSvn>Up4qgMH&c=Rh{UD%RbMIPhcf3Cb4*5!ot^oX&(_*KhCs2l= zqfB4eW}4QRu)|~(JT2!@(d_J%J*zItw_HSg=`^Q#n zc}hp_v7Q8aVb^Y3*Q&U49P;(h$a z#4X{z?H;8XX{EsrDW@}ii-~#K;0^b`I{xGfQHM(*7fyY<2kMb?P(Emj4>dDvqXG}R z!w^Cp%_zITfEWL`TV~C}@Lx^ivGa^Dlb2ZP{jA&ajrxf9;VFO_A}N} zAgGE!sr5MbVFU8feU_c=^hO&XvszplF~v@cFk+3T!l6sKhr-O0(D5{tx!?s}bky#_ z=Mca?9A}d+;h;%>0-DA*V*S1borh~9*XJ$|ewTHtC^i!d@F^L}@824JlgJRJbvS>B zzYFF<7Mk@F|9TX<8`iyhe&D-r**_6-DO?SpLyCiQ{}Gy7YN`x+xV04q56+p8%Et7; z=kJC+BDzH6(R&?-&5yphD*@+&EDB18#4wE*TEeujao$frEG{vz2sOkzvk$CNZsgUE zAMuFcOIY-|&oI?|dG{eq2^=WQ9-;AKxh&BMooNSftz@V<3-HdUxCCKh!_^_3wDkGN zy(J!k35EH+XIrNGjt3-;XeH;={3-IVNbVCE*v!4ujB!$b_{0hilWm~Cs%i}gh2oB8 z%;mBQ&e8>uMc#CL_VIYiI1M@TcVJ>ALw%O^;^Y|_vw&^aRM4rKD&4v7U+x};{ZiqT z*XAZIz1Z}U-0Gu8P#&=_aR1yQ-8Fks7_2b9=ExwfKmP&s$*cF8A4ulKR%7~$6dsso zdQ&t2%&-;dd%1RJn|wHy<@*!swgr$uR^T%l~rmX!JeiFeswAQnR~T z55fEN18!R@E*(#>D?v{FaIpJT+r>7j^yKse+AsnC_e!8a8xP8Hh7}f+ra)DPJ9)%I zte5fh;~aE1`iF^_ zbVTfuI<<*V_j>@ZyUeG<)!cbU8fLylGzGlDokmD}v#Dk2mW5p1^TSkd)P(m)aOKH7e^;2Yj-YJyP_nXE{K^BMX^B}@@i6AUJ zsCNE@YTh`jo^_A1TUAFXG2s^6wKxW+@pmXPmu5^3kV~VFmqR>DJ})lv?ND5XYESb~y+`FxT?pQ^KNSG}~I z0<_Y~CfR>M{>mwSP1QW)H#+BK8Vgf=_mmxrTh`#}Dw`iU6xeI#Y)bq)mHXEkrGudN zlY>cGw`rD%o~y8n!6V^kf=T+83O557=Q5ALB5%FPJo;%hwKVl!>1{dR$+LYRVeWcI zu>WdYMCzJ1^(olWatW~lj7=i(q;aJ=Oo;!b#mY9m0F~-7w9pqqBy7hXW70^fCRhw~ zQXUYc8LbmYhbm2#T$JPaJrKf-r2;LjCLw+{^Tq~w${yu-f<=~rMl`+%84c~J=6I#L zf$ub$QdW8Z<6k4ap>_`9;die=4QbdxPV#4rlUSK#xiM{WPNRsk(GySt<{)PoWZ|_{MQ{r07P! zG^exAP-Z(gF2s@U&k%3sNBI)tGX)|Yv+f>;F+Vo?tNL6Zxcic`SQp-ArdIi?L2p!E z9dQBwqzqOz+9)Q$as#Bcz7{HY0zP#a!gsLaZCGk0)ucRkaP6!n$%4-cpU=*a;#8&d z&2v|A4(*#>E5mU#)UK4y`fs%OYs6|b5`qhLsXN8BHH+xF?7X&M%6eTz&4t>mSNU+{ z0~Kxkb0((ZczMNfOAQm8IJ-Fz;Zx`S@S$z4i!ycg)Ym0eEZjfuK2TlS;yH1d>G{d| zL$-6kF?hn9g~_M*_x}k(UnDo2yM1`B4~yG&sTsAZ;k!mm zYKgq=cHeUfx?Not%kfx0R=+=j4y_J_mXY3(ZTTzHgA6o7iE2GHXQn^*M^#o|e;W?} z2K=hJ3@r#C(Z0<*V;Z0xIIh&*do6XqEnkN<82UErX%M|KmpPuT>X;Fp*mk3vwFS9+pVnwKHMUUe~K+<|?$=beF*cWRpOf`fUZAHHP+i4V~5ykp=Aq zR>wjfq1KS}9NmeZ>r&5ZOy!|#z4VF^ImNiAn~SACmzyE#T|QE45^SJ*NQEOnu?9iZ zhnpSm9~?*Z{+9*dbyILd*-?P?TC!^o;#$}I)7=%oW?@8CkuE?~eT2oIvF=tuPd>(`npI`+@ zq{!>!YN^sBf}xLLop{XoeD;n~o1n$tH!7ziP%h~<7-KWn*J|+1Z2u^F@0Hnu_7Fd3&aop?|qhIN=pwP2}>*^h(^!e4lgFjZ`$Q zc(B86w0rWjqT5pr`+h9<`M;wZ|5%9)nF!n#1u1u+tV$=}6{CS|b-c~r%jUt9S+L%_ zIESSE+Uq6UVE@l_6+3pO3C2(ycjdrrI&aXqx%~}W>^1wsbIww{R!!9e6! zd8f%b(&sh2#tkYUv0?e0*%S(x`3FkKyaT~ML5+s*Dyme5&2%|Exfu?z^tHqD_T%re zKk+{f#rV<=T!O(jFj!D{v`YQ8O`dWG1xfVe@8LMT4ajNb;5Xj`qQCggM-P`?s5Q$p zcw~_xA=IA=E|sski{8lrQf~2BzQ21=-S3&{+0wZ&HuOx14g_8D+cZH)kyKZBzOa=g z7MoqkvLA(Bi9g=Z!g9r$rZ9XZr>)r&)YOq9bi+7cyXKRs{E3#qBGihM^}bxr_3OlW zBeb!{7yK<09J~2z4jqPdl0{7i5MdN`V0anY54gzig7b2ka(n_&pihsVO%@m5aRppdlv0S3z@Zg#X8hdSka4&UyioA;|rE& zvSMLbSl8K|w-F2e5^q)u!*mlPOEMIkdklWUPe+aI#53-0jtP2(O}DFbgYE{b+Q9_D z0jx&{tVzA|W2++?8#yCAtaVt?2dA9ji24sAA0+o2XEHv4r>^n{dRfAxAq40y9inpA zvQ_`0-{4@BoDT!8^Q8_OtX7((?;N(B+eGL-=E!q>ZTewR57#$phME^BvV_s|WO{^i z*}AZ?eAFf~f92_}A8qE!z|@k5`+9iI&MVKXbDm44SIza@^CmnPvfyy?K{!LUtliJM zm_GJ$CntaJP@fpPqk@zhd?NBM)sD+_1BuX=A>Im>xbhkPy6}u(o)Ru^w%2d8UM4dm z1N#fG6#Al*?aU#Znzbj=wG&enho!E4T2v_Y9uH}EQ9vBlJf=(|v5{K2L#6qS2D@qj zN>-~fOSl_lSYse30;He$vna$PHUnO|>r)^O&MpYIqR4Qj-9e1`ft?T~5|AuJ60zk^PP9ct`vq?#RtkmtR&ro_e1aR1it z;LQzgn5IJc5HNt`C^_rf!56awC0qcwXhT5X5TR~jU860Us3y=UX21XawQ9=X8kTQn zSvFFcerbu2E>lGv%bWNd(9R5j?@D~1B#1M4uI~2(*>rj{jQH2Xl8~>bU5vVCi4SPB z%=yfzGg`h(95Wiuw>rMHl5(unre~!d(&w%@UQ2DNiioA43Sd`N=6xBs0O_i59h6DC z%I;-$)~bhEH|ENEEQz&L0N^xUCg} zS`^)BMakjv`7bOBK?i0$iYP2Wzx`-0=Vdn~?j8-VbI!%>si-FnKAW^kDegpZr5Rw( z+zE#Tb#ha0+3Y!xc@?GF4GWGrz+J4~rpRJi3T6KeqU=jS5t)WrezaCN`*X7Y(R!hl zPrsNtY+JRbySfa8b>I8_)sBo9A1**P^)Ua*t!G_&w7yCV{<*trSAaU|QEBRL-NuY= z$%dawj6cWDPnAurfxQyrqNccEc&SW0)9!nL)YBTM-Y2DGevSVHspWPMz@wEPT!!Ca zJZogZ$koOtXLoP%y$pb8rc{+F^@U!&+k#XL77lYSI&i();oUOWTS?xW`lbCr4zgV5 z{&sr-ir75ud}S4SV}0{rWDje3-w3NyNA@vur(uC$=wN@Mw@_Ea-@ixFlk-Dy;x>Y% zArb(Aq6>;EM=fTD$)d(w)&Ed;)O3huJu=MlKQ99@B!6kmDl#?~>zSX96Il_7A2gs< z#laL=$N&g_GC{AM^-lV82AfFO_{zytrrlCU=B}%?gKlxIpmylj$<-zb(CCQ?E?+=OE>`S$vSUWcE=hX z)7C&DUl?NbXE-BQ2O$2MU+Qneh6VR~okqNU46})C{Y>vQLU;uSqgaON+>eDl3gRaN zH%oho86GoM<0`C4ThmFr@QxYfNO)Uc^sk32xNQUdJHDr7fUg$!V83wc&&FL|bR830 zy;JqZ0$maKEWpn&N`{(l=L`8cjpB37%q~c-5Vzf$ zF|j-(wHVwm;QqS`uM&WPK1jR1b=(L-(=WhQ$jNBmrYof;ZnQ*ovfHpC;uTE#lUEtk zN&Hs_CSyT2DuxFpCp}5}dzf{gm}`4{F-7JB{<@@Za_;aSgdNSMsl}uc8P;t^XHX&< zB<3kulRyJ14o_{7c{l4K@*CA|a?dWFs|{#aC`m-B63fB1ge+!P1ZMS{H~*y5Fd4ck zs+m!4Lz@#6q7ZK{hWiEGPce5S&3T>oYMJ@f$x_g7ruBE=#ky=!%qS&VDfCV5L4gv^Ut1 z4=|9V6qAp%8yTE?keG3`8LGAFxe#!Gkyq*dWDW=(&!t8K<{2L$b(4kVr=K|ND;gxw z6+$OE=hw@VkWjqVA9r7Ktdu206{H+?wX^oYaR@Se0(%kEYel@oTSo=!8kGJztQuKy z*6zCn<7)QDyzH~fIf=>d7{GLd7`T`1BlKU7(BCAj^0;uI8A0Iw zw~MVnFUN=Yqw!`LVBUS_{->X&9QJQTUGQYrdGKv#I2c z>3~NaSvV@;_7oonCZ#Y~ajWG)b*mDMHHjXr8dpQ{CtJA9ix_Z;-QrS3Y_*gP! zk_1p%05LDmG8liG80;T8uQ6RrP^&ZtMyr1B99e9;_0EM+sV58q(}@S-R8I^{71h$3 z!}KYKWE7%#y4a>m_eR-`pf@jlyms5{twX}|{h)V7cNO;HH*+XCOuwE9QEmI8T_eBL zLVmt1Ot((6GO}odgST)Qds)ZU#c$}>8EmoKTV z>!e$)u)*(#-iV$9^+apCY>&55i5o1z!IiHi8M^L8+HPv{`-7tz(Sd`zWI+Nui|7fC zN!hhat%Knx@|?}LYkvEqLa=`+F|yBM`8-Iun}TiuX9Mk{yZ^NIqn?J=I&XjFPE!9F z3?H#~FKUfv9ZjMkFHYLS1$LNJL9Y|x?6u>ADlz;a^6Q|$Nca}0B_KZz6%sUL|d zXJ*=@xxbUKc{t9T{+`GZU%fe0^aOw201K2Qx(kC2-K(Q%%EX)Vu*%qelg7;gp@=YY5b$T*Q z-Eac>75dv)Qi%RwxOjsI*iw|(+VfYErb&T)A-f1UkB( z^7_Nl@0pILF|+r0t65NTm~DZ{-@)>0;v8rvm|+-!J`k}!Ph>sX*jv4+hJHDeYL}>g z)5^wo{?;YUFSF>|yU15IiZ+0i0p&8>axTy7lr@dwV=LHK=H`FJ3sK%N+B_TB zy;`Zb=!o~qJGZ3|Utw-OrOl&()x$2{BhcoD-wt@j39xd_h73pLx*v%CzCHsyQ!HDA zl=tt77&SdT79TLE)lc`w_=_(3`(4YBqo>j1kmz?>>m(G|vHmsIZ3x438KArX&U_`; zeh)o0#FQ4ON|xL2``=aYSr-b?4KIG;XN>e@Vu4DE)|-aXy31xCpLTeBpw%Jxtj>@$ zqompA!Lw;c{TRx-PQMX|@3J%;il6L-BVx+QJl~ul`b_Yv*YqD@pJ)xIlMg3&HJ|9a z6xu>+TgDG;UK!ipqJG3EkIU*@JZ}4=pLBV?k$?|Q!OTL#c{+CvrM~rh(MK1zGXp#2 z;y~TdrNd$IaP<0wR7aDVi~v1PLP*bQ(yx36A^Q{C^G~-%ckk}_2O-JI;iiKF+x%`9 z4||GK4oO!R6Hv7PWP5_2Z=GWfgHuxhesfG(5ZP(J+R1C?=cv1T{)JPGGSc)qj-nE5T` zR0GCy{k`scsu3b-ejNd?^MhZ)vIh+rNYqi27>ZH=0!x?H^rV>~%E?Wl)6ZsWZW6YgS}gFt(x6H>&rahWYfc?;v(>H|&x9A!L24 zbfUtLXyI&Zeb4MaQi8xebLOD*)}Q(6%pw|XP=9#6TUH?d0l8eBFo3NJwE66+Q6b;O z5qEwWQ_L4!%ekQYu(?OT>HQ_pxN zkFPVU+q=wkS%wt#J`W#qpP}lrIVIU3XLP>gA`*Y;PjpU&jWjGLef?3U(+2FE2U4TT z>_SlH!#r;WI|y#-?Lklo8ATTKSowZ{mt>XinepvMPs*rDW1?r)w@}1+8HK#lk}8t@ zA)Ug!QvStn8jki8mPjzs+DX9%GNh8RyX%ICy-r;VcnK7=`K!fuXGnFQ#^Rpf`0*`M z_{GmGi6MeL5hIg@sr*id_kcaVh!?-nn{n`%C7e z8g!vR!yGKm1@3djk#BxO+Al_|hPL|yHxSR`jm$4tNog9T;?CZ$q}(g=Y@XSk2D>@q>8<)Z1G=NY7>ja1E?sT@L`Zi$ z`=1X?bTU&gX{xbPNq2ln*Jak?^7l6m`SLdl939!_i!o!$BR-fG*%wn;QpI=;2OyDr zH^p0jTtVw@q3BQ_mQ{o8wPV=^asa#d+WvMN=#_+^F7syxUi#pZ{~Jw)u+9I+BSGMV zQ5$eK$mbCXTIOduzJ!B{u@YBJ$dab3`Aw&7Y3P=!WzxCn)qm3e$4==uAzf1r+46Ir-Ci<;H@zd8fWy1&SdiS=a8Cyr-) zv=zWXz1X<{;y9^7pmm;S%YdB;i6~F}-lq?0??9to)YnwKHf4BD(8Fpv5LZ!wqQ_tB z1fuZ@y=LzMoW6p}@+q9=X~^DwhwO2Pj2B5xL-avXg6dN|RcHW(idsBf00kWdCahYl znG|+x0j8ScIPX4mG$qK;<(DTvzaZfm*XWssH!{2a+RWxD=*0QHoPKO5QZVjWckvqX zJ?qLLz!tGyr9Xs30&|y>#n3bEZjPrJ=QELVcx`p}A~@RVfsPeVG62edg;ts3>3-e8 z<@X0Y?FKqfc>*)vI8}+GM41Udg<>nP z_FkN+ktHqLoCx+G*t+-(+0|<>6hLX!g}?{1-xGu1azh(57?n^x_u7!dfU?czk=_}!P+U_V^7xs*VZfFrJ*seQIQkag#Z499dH zt>K~k^MNg0lm`QKYTlK`<1^oPgHSwui-tlx%Bvf!&s*aWN$iYoUO4V8vGtf;e@s)( zzFF12_Q0Mu^>&9eT1~B5_mMxb%?P}3F{Lj5L1Tte{Ih7&;Wl>mqolXk`Ti{-&<=Y4 zUiuEcYKrK$3Tj8CVfDd1^u^|f!LK62nLSmaQk}E}24dA*a7I7W-a4A6ygu<`@Cqbx zc!%2W)Zjl_Iv81@Dcd-9dzIr`$Mn~dLRp1oQeZwI)M$RYqq^rN-qwKi>3C$v%46lO zte@;9Ssd9f?4Hq^&yF~+(8A#60?hksnr(`ieFy0kCj!GK%BU%d1@aXKp+7`R*Ba05ee~T$S;Z>)m4JHTG`qNK@^SRxL zw66Ecc`bX!gJ`8Ct`QDoqA~w)#$*m}$+#O)%u8>qRAPHtx3%TyTU&l^g06KKG=3oP zks>j%B=1pK{54aiX{i#xDEO}qRzm=l^Oo1z@1_^naBlHO+|YQwZ(k7)e#um^I_OAJ zyD9_^M_N$!+mZVMFv^>>(o^4)s^&V?_(jOySkLmG5IV`QECwh+c6W`H$8CSHztQ9GV>qULec^FixIU zw%y$36`Ka8JbG>z#W|y3j@j*hwDMlRp33dPTBh#HZ-L^@X1C%GK}J0?Pjui_UZd?_ z={w$PoXTdV$#-+M)L6c69wE5PXC{$tsM$y^KNaawW3oPFwJDG_oV11tex#Ocx`7{Q zUBu(EDQ3mTW;#lrY#Uh|&?P72Nz_^bliR89>NRiUeg&L*Gvs{-7z*A~vjBh8&J#*_ znN$l4D3`R5O{^2@F&Wa!2KTbxh`^+Jd*SP^P zl?IhRXpSP7V=Kksvy?FXCCSdlv)p5pQg9i29*4iPiFz&WPV>8xo&LBnrN(qc$b3AY zBsv$jwcsCKEWrq8GO#U{UEVbe@pDUohzQHvl;iOf%aD_Bx0cwHacZGwdJGWRLjTga z@2Dso&w866#8#xFr@Yqw$5wzSslZMD{-{RoQ++MZ9kW)LvbZnxpos9*0P##_;KR;C zE1Y6F6DJ8l`PpY6)?-oUITxLJC5eFYnT>RbW9Xwy#sULyB&LjjcUlm*Z+R1z2UuBjgwU0lk_Z5w5r^t-F@fwp~)xW-=EMQK5+iZFkz!MCLKIY zLj{m{l3VtcGzPqJ&igna$MOZKFsENgZdg~C*ED5hdw8+%hIgi)^AKw(3{0(ito@V_ zqKc;ifZ)aM-p_q6H>XZk#@{aohhz=(C7>c9Ui}Q7kD2u{1ql`0+nl1i8Vr&f3grhw z_?&N&kCfPx&f_z>cgzq!VQL`gQN-B8>&9j%@yqJgQOTV<{wOqai#c)}{=OmTNvqvx zm{h({m}_DFCtN3ytUOR={{G9?ya0%hrOF3?PmW%hqO@jwOU_?B+4N?@cQM&BmW*bs zWAF_VGdXv_)i|S<3U@u*1D|tq4*km=wNT|q*~t&eA_NMPCg1o`f3n$Cz0b}GXH?)- zB}D9V69aP#m%eRA51;leL% zw>2}~XKzcI+q$gmb?R9BzY((lLS_aJhBWYo#ku8Wsa_^cT)_{eCzQe#_hr%|0TGDz!V$}H9f8&A;`d;N!m%PoRcGyfpJFRJ|95hsFb2xDL% za!WG1@JuwO-X+(s7qnoXlE50(Y5&ldLRt$}|Jn|*Z;4*Ur5j|)fR`XxehC}X07@*8 zCDrRSAjHJx{ge&@=}R*{a8MX}-zB;2Hv}r$qr0qkiM-+c#uxCuHIGGfRhj^{7}I?Q z4IxEaeEbD06l(^TvF10=kNl>n?MwDgM{{x4bDD8Y?hXaf;bJ2d+WPM@U)Juc^u>{b zL^__H&v|&AOX@f;k&6PUcam7as7B2s`Zfoem{1YPr|}jq$f>nL>IdG@iH*2F5O-2X zay$iVPO~pP(*BrckxUc_4+N0xKfO&l3!d7#tB%euI9aYqO-sx3Uykhu7NI9NPB5x2 zsqW8=O)-#2#K9p5(f#!*Q{88B5#v2yOpp| zt#kCZ`Q*lt2@Vf?;fb86{*4@Q!Pp4OOd;?MFQqn+%1oI1LUYu=%0()j?3|8sP>(bE zC(s5J%beGiNcVVG25pBbzCvaLg9n&o(g6mxP%Pi?Rcn6gmnH{B-B-cC|I}{fpLdVr z;(7oK5%wkDoLs3fqB%TR^tY>A`(3zYO8-&jp1^sUTp&hJoZ1eTQz? zlVHy4y$V~dOzgs-<%H#cZ z$CiPhW@mX0()@W9_}dP)T(Y^^KHd{61OhNXN*+K)XmP%`9qG!Va>RIg{;czLdhaci zMr7Tyi~5TAaDEJ(@6_KPI?j!XNiXyP#lJLWi0*AZ0y-AqjZp;Oo1SO8*^fvS z{6DLkWw?xlPcvX57+0Yh9{OHEXI{_<8WZ+nUXJfhsOj+vo+ejQJ`|$7AD9_Ok0Cgy z;<;d4g61TuKdW_Ugf5w2WSvNm!$nkFYi-4cZ1wxgGm5YFiC0(RTmjTt(F6e_bAesdU$3ndsoeV-K<#t`acKHs9T68tTzAeztP@fL)v$$ZGS`)Dn5 zIBFe35uk!XGSf*HFTt|9HAz29w(wC0)6g0^SZ_1;mcwyD-Ux9zST8F5p1fGQM)#5!7K! z3rqWt9&$`?7ZO&TDd_oVdZN~FH|a<=xMFHXJXLLjxf!F(iN#-h zfNECG?9sl`^7I+OUL{8>B#{rtrc*LVV-vVSh@Wa?-tG5;9kDHpM?_k0fFIdaF^~9)et{3Q ztpU|JtXh?CPeq5SP8j*q)7M}Qt3>hqqXr(=misfF5w=HY{$aujW(APygs6lL>!g1> zVI{|965A|-m4>`lGDvP&9nxH|dPFuex>JlJ_pT}D>-xl|6^}yHzmFu?V)pj_il&Ln z)OV>MIoUldb~2^{IyFYNgQ*FsYKd|qWcy@^0mC;+!(HWj%s*4X*s6L~vEae3zUOT; zXyqLhwuCUXAfiFz))0k{>QEd%jTQ^ET>dDV@HL5n@88(~c&g!yv51Qkq$_>17ook< zAey&yLjm^oAV9ghA&%`^)-I}Kf08aJP>R;MpCO{j5NsmtuIMUOmo(BLv{gZr`Q?056cQcdr#7WE@9fc1IQ zb*5IW88?2R&-atxzwd`;>;hHJnWC(>gO5??$4~&96t^&_{`og<9gi?Ihb^uzKx7`U z5D1DlS@QoZ5nv)lMHmPaS3<=PQ2iBhTsAV_CHT{Pbn=jDQy(U^#Mk^pOHaRx$f0@o zHYx*Cl>DMj;F+47K|7{1ZwpM z@6CicvqWzcBQ(oeugP@b>gVg#fPEZWqTaqZNhK^m?_TTe{%4VdX`O7^_iLb>=x?uW zp2I8is{)$$6n$S9cgu?XQ#R44Fza5r6O8H^qNCQocgD8UnuRO*Gf!Ncn(@A^+d9p5 zV1cKd(J{S2g~Mve4%gRWGCccE|L{j!rjfs1zrWj9err>8x;n{=i5MR*53pZ!epdeR zR{%R>wSIkL{RH0XK!wbvr1BkGtVCq;BZ>KV3z{y$-RjO2ZbY5KT<>f9C4#aX$%2*b z_4-Ip76E0B5cV(BPn1jzqs8r&Z}EvoC_7<|%xNarAUjX|^l$K2AT5-F9KEar_DVpj z;D;2tpBO+!*~k;hY1Oq$ofg2{t2z?I@h~BJ<$|u#qIFAeu>?I|$r(Y&yQxWH+RO4< zyTL*6%$BQ#!K!O45x$=tw4ZS3%gqULhypW>sg)Bg>i`zmoqmv(H4y!f?l&7cq|uFW zIHKaUR%Hr|$A!rx?}1zMB?XT%As5>-o2h)w*)T~Hoy9%@FN~Fi?&1q2k+8Fc*A1u_ z48%5$!f_Vm$qxmR1bzkk55M@6@h{!G_-+rh)VW{!mMJ8DmK$iK*bM4Wvy>M&=Gl6V{JQpU0enAKY&kP|6V&wui^*w%Y z;J!V*TmmMNqHgNhA|j|Ir#c#Oa2KkVl~LzZPt^yl0t~u)v-T=jmXFJieUfsx>3FHdY)SI z??3*>fj|Vs`u-N7uU<~T*8|FS&zF$^Z*;>fd&7F&R$~OcHN5kcm8zUao1Yrz7GZrA z!g@=40H|+%L8MGhy^rq1arnreWF~{6M_f4~n|@Y&Dsr4A68Qr@JwjsHpeJS5N;;Lt+Dg`ESjF?hP-#>4lVoz4$jA39ygd4zCo9((zmIOKc=Fv zKaCOY0X3#R?1W)w3g9ob$J(Gp47q%;Xv*j!qI^i z9ikRWE2wuo{ehX^fRJ3abg8g(Io!R_7~Mav03S(T90!raG}T^X12JecZk15n87 zCP91&np+RTLN>+kZ_2Lk;lv}Cjjw5f#15W>cHf9wZ`PSg3}QjTrM^v9O^!`Bz7}l{ z=gWCn4?C}3x?_aV)b8$bNn(WgOcSo(wIOTbuG|#_NQBtCyv8hqcogtMel^NwQHW4^ z*zBNBt)lV+Wy|z@NQZX(;fIoLKq}{|rsW!(1RE0_OKM{oMYRiOR`J$>X9f%)=HKgi ztAO?QY=q0ZTl~y@{BoZ@bYqAb z^GLw>+e`W3EtLBdFpaEiUr2nqniYo&42wfr@IL{RokxMPO~P346AGpL;w?Lfo`Z_osjlw^tt?fK3 z%+zjyHhJ5?q?S*Ps#WzRN=BLU=YM-25qk=I$EmN*k2@SbpXNTRKpU|gG;sB>o8iz? zAizLFoiJT3tUYC4oda*^I>%Q&;arx=xEj6dR_ShW4y5@)6YsZtUTH_gXwQy^g<6o2 zMf{Hy0_>23&Iq6{Fi}8`FP1MJ!bl=R7ee5m%738--fVw^8ol!mW4AVTRMhrN(WuO4 zEe%G&JjR`CwkX5~yqU3RYwmFV2ItS5H>KaZGpy-y#dvgaJMwUh&NikGE~qy6-^w~& z!AKh9R8W~`yR^o+tsdv&6N4gBZ*em1SfLi2dvUfR-1$!{g}R2 zOa%=SF-+hiul)YCnTUOK)l#eo(Sq#?@9)4Udl z^!#){V~!x(uqLHgD;>#0oJWs`e-x2xGdA!9pTi*V0o#_;aO>$0@_gh?5GR&_r7_L= zyIe-n z$=3{JX_$m-tgFlY2B7mYU<)pwv`;~h&|aUcFg~$)uH?UPmj$=8)$0AU--E)S9ZnyA%*tp%{%|6xd!^-U&?Zs1{|s;c0ED)<1{?> zlZ+D=y6JD^n}ciHY5o1_=OZ1LrCsJ3K3!IqZ1si-1^6ybaxRy0M1`Mj4oMeYX@c6+ zT#T7z<3@9p`BAL|TnuI*qBVG|cQku8!(1{eR1JWTUooSnM%8g(bFMqaMwBCsu}>}N zQsB%tdj&kBtlZ8$tSM*3{g^N#u=fKjUMM5VukBe$;Y7c-aVsGq9+cc$ zJRMMn?{JRRU*ljA@=g$3J6WxD#iQrEMEXEvnC{sWFZHCnm+|)+K22o9 zw#|Oozb3^5dh*%#rcU&K$}U;>Y%+@VuSuXc6oMJlB^CLdy~(*~F(UQugjisP%;M$3 z-Jy_D&~z!PjQuesbvH@SD4D-p{^~C1+U#1jkEX zr_R_`Fn>Fc7NbpW*HJ?Z{P+KPlrKa@LX|hKwf}%Qha%0Ep6?oWTYrpa>QoMpl!H3s z>$owKZpsu11;^BR=cg|B)Vj!z-t1Ewd}B(lwfI}eSxykIU+jlBW9j2)21=L{$mfcg zykh$Gb+D8k+iHoqcG5RO?xL8$(FkBzCDGS-ae2N3SYpBZaUYjQ#}m6r5ly=h@Fv~c zi8fdWufk|)1LuU#SD)7zg;!|D9XBq-)>|kS3}r5T^I*W?xt@~d0fpcOoXh(oV#SJM zST83t?@vl>fMb0Z(}ji4sO6&pe_VfAaDpVAml;39KW|6$Bz)7utyh3VWp#^`Q8*Ex ze9Lu{5n}H>|UZoqLFWisp8Rr7q=3leP5KU5uK^GV?+>uV$VjrEapKL zmHWdosCJe@gi{lb-r4cLG8gu8^zL5nVmtlAz`reYhq}(DR!k@EbitkSH3^Cx407Ht zgJ`{30YT!Q_4)RZ&yxC;r^zd|6W}WQuhRe-@-BAamhdW|dAI`R%jQmw`~5;wNxd!P zv}XC8VXrqfRAQy^5_A;%<`+0c66e1^mwyi#i#PG_wxaXy&YP+X_hZ**Pd)t^h7a-G zdu53hj~?uhra3fX>aY`M3Df@-BEWI3_V*(=wi>oXyJTo^urx}@yLo>WVIYeU7(s{} z*|v;7b!gd>*{tV@ph(^pODx9dt}K@gyWOfje>6J%gES;FJxuUZ`~Kk-`2BgK;UgEJ zs>w|U>hlf9#^C8d`}H?CS2U=_)Y*l;3@Wrq^N{?DCuZMnu*uI2-AKGN30unS9)C=p z13=X!12y6Lx_AClJ&C-E(i0bUl?xYMnd~-c{sNVl}cPgI# zQ}SK@k7O#Dh(XuhPIJ+HgWz(nWW+ASZEq0Y=86C>&O% zYZly|-JP#tfuv{;F^rDpEJ%e%Og%mEY}J?%3rHwqA%@-dBGfa$`W7!7I;juz+R}OM zV!u$+J_Qj0tn(<QxZ`m#7xt2W}uj8!{nJsG$<%;s@8nLwwhzn_a|X zw?M?$ua6dH*3%N29KkeD%wzE8ulcISm;wNNk%-mv$;uNuWeN0}4^Yv};R3_v%Cgjs zPsJg&h>5->Zp$XeS6#P5@x5q-h@ivT+5fpk9^n0tkLQiuP?k%bW5o=8qblrH_$;Zx z{U;g}O4xm~S&TzRiu81!D_=e+NorRx{M@55T58gJ1T26(f0$u8~=Wghc&4dxdlzo1);gEHt@E z6F~)7D@2(P+RM6}trs2L8@n{aeI`gtRNZ7_G7oK$*q9E0<#`%mXmJW&bW)i8RI-$x zLbJP3Yp#(%&_ZJUG+lewDtx4cgDbEj;Nk(wAL2ag8&v?tdz4uorH!W_Z@sNiA0Lg8 zumPqKo6}h&h`VjW4Z$bn%kc8T(9D-a`ayoa?t z`i!`Af5Az8%%rV^>0%^I!}gE{y1p(Dk_)}UOD^hrwPaTO2}P7%M^asqOt3e2FvhLU zzknK0))%Ags@Q(`N2Uh1%889Vzh+-q%WBF_B4t2A?A_oTSbm6d2m^$1>nyT` zcM696J!Jo>@G<<6z#O$-S$z0xa(2ssp&xLN|N8$n{dx{9b2nJF`&@%LdgfHX!Nady zV>2;FN*Q^1^b>Mz(-IHM9s~6xIK_fT-IiEn{7g+D?Fv-N+JZx-$#RS6N8Zbk7bzX$ z*B!mC*8;lj%z_Feo!O0rJB$b=7+M_>9h-_@_m>*RMxfSRnhGl%_Fj>Fsxo9vy54Ks z-AtuRzqZjQ6w3aPtF7hBqPjs?<@F_z27}E~e95>qxZs!vayD&fG0!MGK;C@!-c*Q+L&S>N=MPd-~OVG#p`& zG5$W}U#Vt@idZ)LBhk&uLj)06uQ4zLu3i*c^3 zsfuJM1Rj5}msNIyBVBeH8xPggql*Y1=q9s3#YpC?%-gc#g&!T0Mb1x6H9Z|ux=t7x z8|=5WDQ`TLcqxU{saCn~*g>Y>#=^bRw@LI#gB}WgP@Z7U%ufxx==m{m3t;%jI7BW< zANJx{8sE3g4O*l1^Gc9e>VKLPjKBL!1WQFX(yV005Iy|kI|kH0ebGUFf5SKH_L0-1 zTTf*jlKh@BX0}AK!AcRshh>1CeJgMg=(Er-3D!ozR4An-qg=uWpQIunhSw_m{Z;Pkm4W12y=3KbVFlG3DXAyc^5tonj+ravA5e?Q#oh!P5~AE=}_Lx1-Gk}$BH&uuu+~2 z)|pI^8MpV7Mr8Em+TR}zd_j;L8B)iUH~RL86JeIKQCu=F;PBQw?Fjy+)-u|lv0RTF z@Eicy-y_3MgFP#0@=D+~d#7!nKV?yTVmJuD58~3mq8Fl_65Qtlqt5y+ zchFb+)n1!^qCEw-0*J5c3&d0ui{B`g7Yodkav8lzh!8*F#7%#lP7_3B6=ix;xSeXi zS9sf$7D{klT-?OQ|APHT>D*_xVI5~YLb$}_4BBud=P@0vD<_OlL7+U7v_wqWi=@$* z!IX#L%Jz08Wp^@gwR(%ANG}fY_EM`Q^EGYo*Wvqs?1BL3VOt1}OW?sknl*!6>#(z@ zb|SlA3Hmdz&xM+K;iq?!*s`17FUXY9LjCLeFjCkP`%T<9^hc*|85r8Pb~6iPu+n!G z+e}a6<6-T*wMcu{bI1r&yJYeEcBeon$KlQCbWyKoX2KWItFt$c@IFlW(_u{GP+i$& z8rV-l6fVmah@sU2Yk3*;4BdcvDEqCow`y^bI^$o>y$XFp7DD*lT`U*VCk~jgSI7jX z(KA+gQ{x=F3g`+G_)K$Nwx1}@xpDv7=hh!Xu>sF!&`TBx>P^Crwm)TvY>aE^%3`Ig z>7g?5x&JH~@B19r*O9$>u3WiGm;grczq6G8koSAMVeY_D$_&C&)*V@Yr3!*StL z;;6AMb=Aazw63X7mt{!P{NQ57@O_ql@sNB{vE%dXC)+VovrjdHw0AF5@qQw!V;Ma2 z?cC#vnW?$XEpo|h)Q(@%1jK{MJf3b)TjQ}|0?;5)yMF^w6nGqH#8aFl+U4B7JTNRb zDhFQFXA+BD07Uiym@58~-#0tWQ8o8MHOs@(L3=1ed&Hr`WnN?VmuZ3swbJBoD5gIV zf$#gy>Cc`tnV}CPknC)VCFnU1-lbFfh}qL#F(vy$@PXD{4OUIB%cZ7dd&SanG%VHMd|ig zuLt)^X@?G{N>4OAa;Hj^io`$T`F`8zPjYA;%ZPb=bRUDXVn%SNT+(PbI9LdQ9!72o zK%Wq&H(WQ&8~ZrmVOFAzMg8{lNB#2sYAItUeyyh$30or-r2lw7V%?WlSBDW2N2zzQ z}r;G%y@$upEllmWnjyhSvjAR*co*$DrkeTnKQ(bM?o~ttVn7TRpi5JBttpT z(Tq!NN{Z@H8jqF(^)#!xtEI~Qge!)Ts_Eqp>|6DI1o03<9WoW`LO8A{xv6O;0~ljd|V4_5~$MXOcX*q>W+ znzZppqOcn(QI%rg&`067|Hsu;24xj(ZA!Ykq&uX$yGuF*q)Sq|k(Tc6Qc^^&iGx_ z_-Q}XG;IS9dSa7k$Y|#)AciOK&?%0+9Me~x8fS%{+r0~ zxoLUPl=cXb*=!In0-p1=&oh5VfLx#R?YJkm+U5u6bn`1qD9}-AtzRmyyA`aI5my#$ zv=1jNk87f0#TNa^93?kC^=FYNjs+6qT9^j2>kwb`1 z|9Ei`d{X)tDGA{bBJyNm81CG0vO#=hA=~fq*Hf634dnfOkdK4DPeHb&LP)K z^R|*14QqDsRU$`y{LdSv(=!HUV)J633^;_*KT1DcG-8In!r%s;@I#Lb!j_ASK@^b`NqCt41Yw2TP5RH$bVT!$0T&l%xh41eW)>#+ZjnUcSQGIA6owxW*+YVI#8 zmGmKe^dP?HmW}zYS>B6>xsiyVm0w|T6Ykmv-S@7QM?R;idVkAM*C=^Irx~fS%r;jJ z>*j%E`1hN?=Hfs=4U52D^J%ihRVkqlcdhd6x#wC0Ty*mam9WzhF`frq^XfLKm0>Vv zqPT7#X)A=eb7nI7tDF3PbU0*d!ceLHw~%ld1ny6fEc&7SuqaE_vhs*fN`%yZBS;*EjWe!4EBt0g}D=H1JJcMtb*{r&b6QyE%dp zm7W)bs`N7V8qL#4FG$>210O@zaZ?e2Xw$e5zXulxVw+}kA!yH7@<`Mfj&BhNx)ebH zJm_n8XhiY6?Tt?1JId2-6G5&AYd6!W+t4oD@9_mQh|>A=We4F~reRh-6~sR$zK=ax zdAACoK0E)u*M+RjozUB~%IW%0$(w~x0JfK0rU*TE7z6o?_&T8Tcs|D9+%Au}+EtPfhk;>WB2JTt9`OlQiSg1vIlY5WB!k-x6p0%>-H53w1hI4vESoG zz1-EVXvbSJ-2*mGpk%B+H833+Ia_rN-a#)jRz5j2LX9c%TyG_Ai}L9$wn_O~24Pk= z)PBVEuGVQbLef{~kYDUK5vs0q)IoUHcpu=#8eb$h?UX!Y$uhN#7M7r2+c%eDC5{&u zyXXD8{O3#G&5jr%DZfCVpykXL8ndNaa{;oic+K2I^n#hD{cG$G*W^kQ1s2Qj)RgTZBcQ}NAwzPdoAM;A9ns660#T=Ef>v8oP?kmgn+&`LGvAlU-~xo{@3 z724nq-F_SyFAf=RuzClrvugH4oF=7QV*yuZe=_j;m#HhRpO-jK*TOO+k)3jBps|ZG zxlM$94bG7_=ho}VH@TED?3c%OiHg{S2`_Fc?+Mz&LfUvyJ{64++%5CFODmlmy~r>@ zt~~dEX1a$1M{F;YV(P4nmo|{L3eV&&c%4gVyJ=T_pm3|NZq1`FyGe41LwFWRS>)`? z8~L`WiaaU(Fk%m&ov7T^5s0w8ET!GdSCh$~ZZCI1JV(Hv>w2$NhvMrA< z9i7;Bk2>pNi8oPBDur+>W@_IJGq^JFVnVV%`EwrMxaAoCNohGhmC?U?*S{?+83UM$>Ih*Mzo4 z4t$Urd3Qem{8g>(fS(Z~$MdkqYSxBc}^ROb`P(kz+9uJ*HXnZXeOk~Zlp?ocdjV@^L)WOZ1yxX zK^Lzcu3~W;T%SV_cDQGq>omfV198gn#5>D^8fB!ELbRG8F;O7cQ||w(BQg6*CLpXg z<0iuYa<;=hhlLNoo6zs!ADz&xgc8%HB$?jJ!BWVwJJq>TYgQS$gN#J7?#8WJhK|!2 zd%Jg$XeoNM62I-G7wb4oPHPFH)#PD;ku+R|TOK`xIsI(26=B{_9a!`#Uy>PrjS2jnujom0INR*ZG^a6ElfPNpKCf*#`&Y@2zr0m)c zzS70#%Ttvt67>k>IB6YwShE0bmJVcoL{{^7qRXaROA#mb)bGxy-!u<}(96z1!r-*g_EZ1wy=A#}uVN=&Gd zlomfMlE>{N<>wXJui~<3o5;E-)Gk;@w0+#|@w{&|(o<`yPkeKJkrJZ^wDUotykuZ~ zZqj#fuum1uAGM>KFQLJYF{0lLz8e%CZ`OR|<@#L4_Jvc?=UmI@K14qN*N86jP9;6P z;>bd{nmOyKGzsZnjsCy-eH0i#rT{DZp2`mW_6A=y8N@k{ga!F7<~Jy7?qIoD3qx^1 z3##4kS9UA3e#wqF;*aOB3*kqJubY~PyeyQ>GZLa#ZS{sMSgVvYggo82;A6x?i{a@~aRj<-F%iS?(ERbIbqgMQM+QFm$% zbnld}6#oN6{q2zkAJ(|2-yv+mZLbl->nRDtz|N-#ieNayFh8YKwc>TWEpQ^{5oSf> za4&MP%qp#ky-xTVRe7OC=giS8PPH)XfbVi)1|{s5agEWr^TQNIj59?PJp5;Bw?wq# zK4rJyb0+|#k5xK{Cua2D5)*_CTx((foG!nQpXjRfHQI5{leGv5xiyQG(=jk2Q;56I zh0<3c2pdZ50dActN$lIqw2aBHxdd0-DEeQQN?5W*LMkujE~a|_+-vc{tX^_^WfuXk z3fU!w@z)kqBP>ss);k@(pQlAy3md7P^uD9%Y^IJN2cx!gzwd`XOl7lKa@kqfkK!^01In{nj4Tv_<)ZV3&a^s(LR(M(HOB1?-MBNX<8DEVx8fnOObYrksn(9bR8B7Y8<`e2Ox{;o5>>**NaLiV=CLyW1@ukj!;axe)~D@e?Jr5^ zJWy5K@ODr1Dw~2d=+v}@GgmTpddy5j-ghRonEL*86O11=ZhR5U=WdgDj5y*-sfe7c zMKG%D6B?gsNs2mU3Yl7>6;Dtc(0EsZ8i0Q^>;gGDi2F1ri)tBKmI`6abWRDs$MdcRY%u6vC)|-1od;hn zZ%1hemtK>mTz&-@6tkI<>vxMunYumNBUY>PUhd3%2wjft$d~+r2ReHP;*kBf;vCn5 z%7*|%&k^~5hDiUrd_uhQIeUZZzHb&#u-W1&Yn1I)B?S}6Dk0ineYLJGn(LJV{o{gD z_1^c%*LvV~yEn`$Tl2jg{y-=oOKw`FBo}UuP1clmqm8vloZ0p6f+s|-ewr{E{0y?k?8>qtDAbfM#psHgT+VwDP76%{t{ zB1#~knp^TfF&R1;j64Rp`Himc9~KeukM&($7zKW7VsRVIkYgAc=T0tp+MhI)ja_~H zA$luVPb3Z^!NeO)HQ{}Z+07}}LH%!{)?FJ77R`BrV06SQ*wy7WUqgH2w@cpdO zkCO=6q&GYry0UIa!N znK*o^P?`YNaD9@xs9({}3-F>EMpvz*RRWj@m3w^X0960o-wZSsX_o)A)4%wjFmO*#c#9wrss(NGunx&WX}Zv#Nqv=K1VT`~{AQZ#a+>8x*krHX+QFoQ+_`qYCH`|VI8A=vqepj z=x@OWmtu$lh$)pmg2{}M zC4&`YcuLk%k}x9eJDF`}{XtQsxn)UsFhNK_DxWPYq%q59hL0+4C)|+t(<&`^EeB?A zIu^|!DPL%F$t=fmzQ6NO>Vsjqw%?I&nSc_itL3WzCQ1Jd3>e3P11&}z@8H;z>k>7Z zo%5Hhoi%~<)nEcB3s~Q;El|;h9yEZ$$1sK_DT`cJEnWtU!j|37>W( zcOHj^28+0|8qgRPkW2LV=$I5aTSX3__c1oW%}VSG_+V9iw?4V|Snm^w+2KbIn^8k( z=QeX;H%4E3w#IKEO>VrqI5mc-3vOda!+}f3Xl=Bknf6&;Kfn{bA{7aV{;?u*f5R zRe6wDFq|`Fms%0%O5VftR+pSwtr7~nEGmHiP)B)9z>q*%ZYq@z8xa)KWmJ>qd$y(n zB*3?utQ4L0&+ocvIqJnl{lDUJ=_$hbE##9Rhxy;LMQoB?i+7S=9xjiKdZAirYc7{G zHo)~G7VW~DR!%KK*ggGq70nL;<9Ka5&NXg54EJ5(w<-EgqxPr&0!5YbH4Y&9*&Z(q zF4npqMFu6(&|moE8z}Z_f;CE0vyyq%(%sGohO}nrRQk0D}HzNp>XxvIm_NjEgfENO3(S5pM8{O-<~)L z@DeObvo~R?E+4L&7p_FiJo-bqDSict>Qa1Xix@FPoY)47ol61_0hFtgM#yR=2xSAO zl}zONbgpx`fLwLrHj44uN@KAlelLyLS6VK)#L)u(hcYUAOnOuAc#6>Vq5{`RxJElM zfIo)pSpc|NK|)$~LoVK?vyOf&^oSN#;5U|F|0BValuOD@yh*$OU$LN)KPAi+^Fx5| zL%Qs!wq!+EK;6zP!G?i{iv!n>!j~_^M!#v`@JA)LDs{tEM$wTr1b=SYaADormwTaa zokQr~mtP1@4FUPIQ47Hq>&VV_SiJXJtPKgqeu`!ItU*>g`m6bf!IFjnRlhro5{wXw5xC5$$Q4J8bi&cB#!Fu zs-t=2;)d}a{!jT5y`^w8?L$3eQtyQjZU15V%8vWjy)1Sy5HKf_(PhAk_gfi1SWK+mQPE2ueY-$jPfL5o~$ET z-om&?JNv15>)LL|CHrR#;{x@uknIQDYszY)D3DkKETU!IEbO({>m~UuU%`Ro84e@5 zZex`)9s>f7+Lu19%RbWhJvqJ4*g!fgHss2e5x@=Xab>d#yQK-~N$W~{Z|Wfad1m@{ z(fRr~`jOOk(;nx&GI|Xa)b|@@H`SV8kZ^b5G6flP}7cP7R?IjBc{V&1!W z5z&1E+6VKjNo6vcrKdzz{C=Grligw%?^HiF3Im32G)i>CW_LrPLVD-n=?Y9xFTYeNp)zY-{RY+~-BYeb&eq~kYGAdKWYT-MbR<-0n;Ear z(UO>^&BxG`pswdsff7*aMQur$GQ-S}LXtS2TJ;JMOWSsSX=|dft5j2PsEC0`gvrT* zsv)mq97w{&U&h`QU!h27W;on={1k%pXNKIQbwTcIYY0QPIwzZk^PR&aZ{w85Tk2Xd zXl$dO!)nadVq}NtYrbD8PU1|%k#bOkJeNKcwRjWpsFz3uHeKLP3k#6si!^}JchfB! z!g4Gn5+^@KLdfLm?XHD~-0Z3#Lpz(ntE(~z9%^?K3(ZdNqbl1;*7SD<1xws?N2>p4t_?6+04ikhG__dqEW)tkEGVE_d7K$0py$ zru+=fdI;5RFyxmNqsRzS~H6ku-z45-IU#t2F{>9}IpQnEhsoQJo~&B>GpSFv6zasu5FAiK07ZRQ!I z9a6O_!xa_ou}Jaey8VeA=KJT=;rjjK%_@fOYHTLOb@apOy~k=hFGnao2=#nzrczdW zCRW5A^+@6U>($zO6zZ)dv@IHKrRAQw?U-;?Y_Sa_B7z>0adoC5$$<2VB%hh#J?9K% zh>cLnOBW6h3we!Q!&MkGf;9SyN_RRG$;kOo3q^bpq$g5;0SA?ES^>uYdBS`$QACJ&I{!AN{OST3 z*(Yhk?`3SUFzFsR58ze!3O>n|0}DtQmT&sk=2illUj+c)aFmQW0eR-702cH#Yh1$Y0N!UMVs<8rMgPKO-(ngI?o(Vkr^HpZEJ ziU>!swFZa4SnUIu2hO8)zeQa@?q$ABO>dt?+dx&tq~AL_FM1GU)ZE@dPS6c#;XL_r z!tkh&J|$Cykd=HN({52PlXcw-Z4fuRv#E%=N+Ey07&GO?>`h{dW(MGlA{fT+ca165 zqd)AmOdG^+`T~%0ZNWipAk&XFE48K1tN~^xGKX*IlU{kJb|aa1?xWcp$mPz&PrJZr z4+?;Uyw2;)(RifBkW3mBd#DA`4jnuVBtmNMH6P%UM0WAj@jmbIKT|KlYwKAb`NvSi zhwiJ$e3HUkXRr;H?lf^`nAN95?3XlSia(VOXyf00EB>+l?azV89SFC0HCTThsfsVa zTt1_bUo$o%mGb06wgzAOtgi|xYfd`q>C%j%VI@jb;f)jraKk2G7b|g>^73kX$6BHr zhcXy_T#G~G9sC2&5VS5u{F>_xuMBf5^Xd0VJ?l9Vc#J!Fa{$I?8MRDbBl&=6+0a?n z*L6$Yd<&=A@QnlUoA>VwW^tR-Kb$D?`aqjnY7X&P`lkNRjjo%#rcjZf`*<$Dm9M}? zGwJ@p0ibgSi8hhNJ#sBPsv&dJQ&DqlOm)1r#W>NPat+~3ZmWhccKqw0&I~m$^7)KL z6gQBXK*I=<$oZnde&Ct+0Y_7L&DGud?Z-#bV1!C~*1e5o8Za3<6e8Zd;_8=3QhXVK zLgw-I47FEuTN~a{w$ePTeZIZxykP~t{qbt`fyIwvh4?eI z+Fn%hh!VCV??w9+Nw+0M&*v1S%cY9T{0pslq?Oz@ zx+EX@3ukd3_C6Xb=QO#2^`mVEE%#>aPqij4oqp=8=DtS}jP*_;lB+5|4O@5vJi}~9ksz#pyr}T=`Ur!|G_*|~NGquk14k7lVQ)-L z3Sk@i6cCCVmT!zxtzhOnw4HHayzz6dmPX~6NSuM2cd2dd43!0xnKw+2)FVT=jreza zCxxQ1o?uu#-bMs(hB?fD9&df%W`3hrhS71jsFaGxtCU-;YbEhe>&y&Kg%h}|lXv3| zzPzfk9n2-Tjf}dOlV#D7l=IQ6g&`pIX3_~=(-J`wx*>iynp4S0q*jb5|8xrlz>VykO5y>N-isIajk%r=YE&E!)`@QWJa%o}@kWVe%e>qFzx?ZRSz# zcdPwl@Yl+*3_o>I+m8fv!kdr#rTeU6-`HkF-bP@mV zgQMTRhWU;B;}stJ?}~53NWlq}&hz}U>8D;iecB{EVx14r6>a)TdFLcp(ql^zrI|_%e_94KK$v!K*GnVv!OBX z8qhAs+?gm><1x~?x0i@cEp%4Khdl$% zz5L@}%66JE0bSl6_ljDB-^177(6{FK`=@(0B2H2qWnE_!2YAx2ZXZ&oPHX#@!o`X^ zqrHt3l(6EO7kB0-v{Lx0#CU1THb`>5AYQU;zM3(ZV8wttt5UiA^)B<8ISjzlKC5eR(uWXJyG=K7Spxejd{G_U3E zj$Mvi2jbpA?ErqMzuotWcmU63xj)RK_poa;eSeM9Nckp4QQkqV7Vv5M%XibBZ7}J0 z3XO`aacmJ6OA2rG#v@WM;k--}3aRZ=eXaMh#88&KF+FFqMzI}yr~@@@MsKsKW(gbh z-Mg;U!N(k*<7Tr|QDnKBog(LmYB zUJTf{MlZsun&m9*3p&{FF?muoY{^UWOcR?lG4$O5ejISrNXhAnXec!BHH<87t9CTR zEA|E#!E1FFLOqf|Z2-5J<f;DfUJHHN$}X zDE)L6Y_<)Ow%s{@6e<3h`u%{;LGN@s1mH0hUL z^oc}q+L#)FhO22Si)Ypa8Z#=rxu*9!vfM~6Z?yQekiUz4VLrH$yU7x$b^`4$ixm2O2jMdYk* z)Tb?SFeLsXd(S-SIj&ysnm?)@(nkq}HZuYXg-4H+WZLUs?o2xJm2?zv1~XQy8>bOE zyM6md#XKi^>MQ7wo}|QEp24jb+<96BjGC?B*bDuG8CuwYb#cG zd;zuB0fqCRUM=dgau8avXs0&IJ+Xgh3AXX( zOHkU_dKtqeG;g>4Y({P0+F2MO1&8vuaJa4&r$+2X9dj|bg;pm-2UE?VR0K5Vg)bNj5b;31zl zxtvjy^~(POwayL}Fss(ZJzVLvnen3MzVo@GM(sEaIxdtoYU7u$(g$p7 zUMmMYQ4$>Nfad!jx)SM!Lg z#fLi`ctI-$VInX=F@_WC`Ukb3xkjRU)|T?KeAA;*`6*6TD+p*Gtz7JpEFpn&WZWzlO`CvP5#qC3gpg0u+rHhZE02TPe|u+yL8>fnRqf|WLsVPV z;yADm8u)^z2(Pw`C=_{!lns8l0T#*ZEwcgmUv$ zED{n^4TcX&F_cm6BIIkN6^ormhp%xQ8Y@i87%WYLuo@bbEsWHD{@B=JXA$iR>GKTA z@Dnzlq2GKSUdXFun4?vRt4vg3v*)r=u?dAIqlu6T?zNokowH}%*4#@H(kx!3zGx=i=DosKYb`% zh+`2BNAEPl`>;?n*GO3YgM^3d+Pw4N3e^LtQhW5mS}8fT8Tr=*dA`AQ{e*z zXTyhfatA8698FUCiYM~mZ7Xw)AW=t3@K+F849s;F7pV+^iRajXtX82}w3TMF?*`1z zZAzmoL^<^`^g-g!8@=*QDtVg3(2r7HDJ_Tus+miAqX&5*RPdj3i+%qa-x)*a6>(*V z`MhV+^zAkhC%i&`_Mb0!lK?p@k~?MwubcKk%PpI{`h9&4jsjAL9dUi)B=>+=ug^}^ z;`z>qEK z={}1)agH$!_}U}E3`Rr!%x0e%-W--%4(Nf)n4wYiw$ygKs@kxj&E9U_ zw`XC`#7G=co9F6;t5i9mdPJ;uKuX}2x1Onxeu9+DCTdxbCxyy9j6#(bLb!B5Z~2oP zj{x;65~9yz^gQ%p#l5V=(<_pUqBMES(2j|_SfTXLUL(zb}dnzhjD4fK(F6zU7KdHcbbGOqH%dE8CN$0W_@QM`ZE4OmV^b ztbD(LAEe5OaDln43$~m8ePIO;!XN&naP_|p&t_eZh)@A$O}vz+LLoWc8QbO8?}Pd& zT9<1i8#-jFq5)f$Yk+=H&4Yh|Sx7IMNr4h+yK+pF3^n)A79I+Ip($_s^rRT`oRNf^ z1OvHBetOLUhJQMc0=e^ZPolcct9{2d0kZH=>U6?GS_DKs5ypUS3+$uz=s!CTbUaV) zQN4b?bI2V{53~-Mj11k?Bl~4F!nK;nr;`(c$|F>`c)7%ZjP+A9L{ykQn<%&eHH#{ddB#ilq(2^Z5PAQYy*Hk@Fn5RU>3EO z*%uWq1wS$Lb=xTvwK8poOox! zicB}Hs__o-T;B}bpVCcR4Y6arGB8{B3|n9t7)p)cPQkuRgVw#t$w2a;`2cxM*7NZ> zBL(u@{91D?sCDHg17s=3lo4;Q^8mNB^#H-_lcxv+9zGHj&mKDE#_x@po z06v}2Y0Cvr60 z^-5*SNAW68fO;+L^i}RE81y1Iazz6Lr$lTzC*4!jxW(cfcG>NUcs3APk@W>OP|h~s z_yV(Kv61BTmpX!V^DEmHOpdOAlpz#x%ITVYA^d7!gdNPn-pZgUK3t+S_zTDA2UNW4 z0mDU*)WFvvV!f*mwF`fwEIH^^6d&JYcTOGT5zrLi{Re{xYy-bUJ+wZO9>IoQ>R&e` zwyuEabDjRrDiH}T#wR-r9n;gBE$lkXRyRV=yamJW2Icy)T&3zW^E z?MQ6;?!Wv-P~eLo{px4mm+ECMG;ozdi;;M^%CRXmHFJ>Y#U9vQzhTs|sUq>RFI9Xg zPJeSM-aeQSO5_9r=oOkiOGHDwsE%5r{bv3?iiUfX|MQbmJiIc)0FMQ-bK7$Z10$X5 z0V*_CFW(8jh?n;?thC6RBIR{Z#pbKa^uK|bTf9Id4vRI%zL-$yugv-Y=5b0L8ua=c z#DTqLAuVaR7DrnR{NTCLBQ5}S)Ikb#AHY3emI@a*-a3j6NE)b|Hri9Nq6B|bB=kik z&Z|>e@goK_<0*)3F1%Q|w^J2|YjT{I!XF9Z$<_p5hC$yn$yD0tDXgKTfYH7ajEVh} zcbDW^?3Zn595m4O93V4`x=otnV+&i~c3e8{{|PCYzI2>)SphIE1iKXE^&tWY4g&E8 z4yHcTI^M@|FFyRP6D=)5dJPSBOx0QVf(K2x*BA9nmE38VvyHh&uR>$)rb-{%cF!=H z5CW2SzS{fZL+V#Hh^V*RgHQH?9$+M!EjRs0_@gqsY04co@xMUDHEn=PgdZ?fC_}Ws zH%!NRA;<5|klWAp87#*Efxb&z%2OhdCU(7PwSkYOBC%jgpJ5F<~$;4FdFBOBJ*X^v5!V|1Rn0u z+h}6sd5HxGnd(WUOo0l1!MHlCe1byuQp_%$WT)kCj*!voJIA@O{)pmo7k)tBO)Jw) z{T3VEk&ZcMQ#&6nm)9FHGFFkFAP7FI4Y5A`%_&-u{H;Ts`-gxJm8FOn4x*(izIe2& zUJ6`Gx2*2O$M3Ms59OUxUnq5BCtYS0c`uiDm;+~$hyRx<2QcLi{rBk&?&1wM4ymMi zqskmZOs2wt<`!ZOQWa4Enyb)4I4%WUC8YWpO+IWWCkF|?s_e9{unohtg>Z=e|T zT}dXIj*Rl?+Z%EF)nfQ}zQg+4z~lDxz5Q5<@Fomur@HmshYozH%l*SeC#181s~s}x ze-5#U(QgmX$8BI8JI%(ZObG!m(mbg*VMJd9J^jwji1%ic_#%-TZ1TJJrw2i*s?VNr zvPv8yA})*g0-A8b_NZogCjtLWu@YTFpUYwM&1!$$s=CL;Ki@w~ zleOnH+reeb0&i1GETE&jK^Srl{Wsu^ipI3NTb5i8?5JDBK$L+fi@6?L7U^E;XrE~_ z@IDs@gs<^9*(UX$zj!8a_w(klH&%yXlYA>Fc+L+&F;F84wtHtGa;t3%O)7}v6N9sD z2eU9*OB!*;FXf!)LlQu_{hLWr23yLAjX#)UB~5bVJM?C{3+O_A=_X9 za*_@|UQW@SA2mLo#MRTm>tF|(RTBvGrj$XGym{ASMUFo+&RMO>z4onER1NXuL*UXy zsq#mV3g>IQVv`IK8?z+@QX{_Y&dHw+vq~V3!06gIYjGM;3`KhJo|u4O3nL#eEPY^Gwpa=%GHU&mm!&%e8L^ z29wU{wrCLI0(hX@#;m5@jwSCmdfJW#`7veVVK!JIC?g+z_MfVDwpWNwWP!U}sWTEJOv*I9sl^W+z-WwZu~w?Kd9#%(i&jS< zW@*PL=vs^2AwwG|FghkWz+;oQa0^;v_Aj zvhFF=zJq3;r9ppqX38x+oxe?%%6vLo5VeW7DK(zT9B;IQtR6dk{-6qzj0C~h^M#*J z{q1pF9&GVS(>3Iw=ZE%2=Pp@F$k%`V`V*)_VN5$W_P+5Q!@CcJ3UnS-RK^r}W1N|2 zgLRRL&wjKoZ!PRNxz2R+Fzns0pqeXz3%kp+MU7fGVCgew5SN%W&b1wDXv7%Yq)<@QYz2k*|rJ$3?MC`MXh# zk!dz3PI)a$54N98l%!JGvMbsKoHyNcv8L+haH#P^gmZKF9g1=?+evPHU5xuW&g0>e zK!*jn6W(=!?dkyv+91xy#{I7k0kj7X_B$F#h`Cu51Aejrz$3iDN@H0^Bl0ux56Z~B zF616~fVL3H7@sykKl#cJmUYaY&0 z;3*LQ^>%>;WQT$idB$(`?vHE=+wy&MO2X&=U6_UeFS zf`-Ymg}s0Khl&0g7dcSN2wL6jSLf7cHWP8nwUvwx`pkTzQ>AC! zn^Sr9*@W?q(o_1&C{LKbd(#Gr@+E;NhV8mkmA#Q|U2Ap-t>9^8LNoEyF+t9x|6sT| zMAHC-uWLskuFDyFU(i(=D03;@XF@S^u_-NRMK>Wdl)ds3Q}`R6U0Z@OX~a`J1VY?D zendAh!--fSydSQ)P--S4arzAs8MkOynPi<{Mn7Pm%qW4Y*(U^*a z=sA7sX6I;F<+n}v7lwO4E5sg0F>~8pb+mzw7&_WNC^n>u)@-&)RAjC#Mhw2b|NUT! zMPsbr>Xx#jaXZ4JwwtWYbf6;(MYX29Hn=yX6R*@+@1fK1p*5~uhb{KKeS+vAhE6M~ z)kC!8j?m`QQlB7lef6ITncBysV(z=oo#Wo= z$Dt`^A$wxyc|{nV<2We9tCCOsf_XP)q8%J7hdKLTZFZFep0ABsa^*zVwBzPk&1haG zndvpbFt6MAbti9O&P{y7iisk&=KmF(M3qpGE)7K=9}s#ui?Eg3t?$tbdxqN{Cur$; zCo$krMebomwT2G)ger2H?RyVzL{CRd2-`{yiCII2{fa!G;JReOm?VG)JWPEpJLOMGS1T}2=i5a01XmWN z^p2L8!E5sYCHZxz)p3uME064|qdwO1__82g>;??B3}K@OQ|3(SS-l)z} zN7WrRx!1?(M6gnxdys}00&V7g(nh*9^aGoG3Yujoa^2`lf4W{&vrp;&X*#ZT1Z^Xq z0cWH|CvZCQC152jw38EGdw*v{s1ZAUWh_>1^bhAYf671i8L-<#|c*FZQU;%RwHUo7JzFF5}}MJPnV$S=KV@0!PTd$^2x!Y(dS z-k67Vi3}1C3eW8jobkTcKt8IGOC?8=@T;t9NhlLpAU?(EG$}X{3zX|fGthni9&1g? zbML{b76mRqWk#?;=$`28^6gm;z)atJZ4HxbAL(962!-KiD1T%Eg&fBKniloOR8uO zTNWrw5>bD#zK;6GqEL|r>fj20X#0$yPu(&V@lBmMaHr+PFNv59iXo@~C=h7~S^9uS zh0s*YSDl)e$5rwD-`j6>#G%-fjYmbmmAQAJs_Z?1>I;&Aoy5p^B?avu)c90c!9wVj|IMq57Zv+|pyD?mQ z%!!`LdLkoAJMHWD5}(FV*AINJ0=IHE^4xb=@*W&KDjHAx11YkTb~X2DwxHL)GmKE= zN2lJoY-jdS6g7|qwC!^tpgk@+7V`Tqa)U1BwTUW3rW!JZXDV;rxejYK6m zR2x*n3`d2^I5#c`&+dH(XW{Tu1roP^-ir4{Pb%P%eZAG0rN%MS(;-^xp&v!s^E=75 z7qh2MunMJlL8>&A9677O~cut49;kU38)Kt5zA9V7w$iAUrCpZ4Rg3mZk?woYm;K6}#7^f>^ zd8C+#$4&drgcSNSkxq(7PfN;YeFX~)p_}_;_t7f!{sWxo7w9>2DYf{iaB2##3<`RC z>hMC5nX(E=qZDK4_*x6a^+b_e!o}C)CyGd$M^g7z+Uq9eXkuJdY4M4!#8L7kd!_gO z%4r)-?RpNMR0EmyK6{@b@#lWT(GW>oX626r?g{#-%yqbS4qWMqdkwUCQ|L}2=P4Tl zrhi{*AZg*V3~L}V?VEdkYewWt#NV9tDQyAAN|rhgtGYu$vJ^uh&}bHd@FO?AdiCo0 zUEo0^Bmwms@hL>n%>8WyBgT*FniJ?%egpkWrkd4%Jn!7rDpP(%MLom?h~@2z8h^;* ziaw_pSD#3f_Icfu!MUOpNV&jjI&ULunKfvel%%wnkopFLVuf&@(V{5H0^uoYp4R`5 zske-ZtLeG`p@Cq*Ex~ErCAf#+?iw6|1$UR=1h?P>cXtmi!QI{69p;kv`DSLV`>)sP zzFl?d$lm*$3bJ*$GMx-=NY@kU1OZE`UOOt@d~<816mR?o*EO(Mv%iYQ9O2s4<y$`B=sFW^y^4An)TCqAs)!hWqoHs_z$HxSu|+ zwI!?8R;oEU{QdCVyT3P$Lk4#Wdo*8p--taeL2xZaz?V?qaYSjUu*`qN142Bbm&>Lo zig97INkB%7Stv^;H#{?iX9>#^uiDs0Ni&<(B}FhIJcP#mHPTbSI$W ziW4@#h2C4$EA@j*vx?Wl2;Z8ncrP!Q#eHA(I}B+TY1AoJm~pPCC^*B-W#+?T8o6I7 zzt)&Iu{s$_;K4%VA|uS53?4}|bI6|EnozFvDNm@>YImTt=?e1AcxBQUaxc+Y zdkko5HhgWyr!#(K zf5j7ey*`Co&uM$r$JSG=+l%f}__TehxPCwHXUfcdJT}GiE2No6Wm+MS_X=Px306WL3tchrSq)TGPQ3+M$~MmJ zan}L(##!8Wd!11%47(D(Z%CsU_cd+xJzHDiH+Yri2dEyq2lH+3=fh7vFBerT!=InR zgjYtQ2Z|Dmf#~_)gX#%gS7-KPv%Q}#Gc0!?34<+?z@BRZq;mImOTl1hncj?aEG6 z9tq{@Pf`iuI=Mp$KcsbAIE#+sDo9w%q3eM5Ukr5eq3H|0%lDz)7am|8&~(;H_|>0>Cd4McQurMMqku?RJoAMP+u$eRWZ zR$Oss5(}wyrT(!h!++VoG$;i{5<1`cs4uUuccop+U*WHxK9a-@EH^>sF%%~p?Y!Qk<8hf#pR~iIOY1=BRC5^ zqnQd^#Nqo)i4l49Y|UVs$;sW}jyDtI1LUZgjjvj3XCG%3Z*TKq$Ki+5;OfPR?SO1e z5$61Gxr=lR9=(A$EAbdbAT{dc?_z#@qbfaS43z>d3__uS?fg7m+S#m`a(*lzREw?t zMlS5wM1^M9s!`;yA9Z=stkB@?Z4_r}%6bZ2;vXvliz4cq&ewX{uL-P_7U;m{b&hkb z86ux&%5^7bL@=GMPCNfss3&^n>7QtHeOHUM{2Y%JD@cb%_^~~WXe&&9kq64|^S5fq z=Lhsum};xpaY?W=#p@ZIjiE#{Hk}%9G*yJ2dczZ#E*uX7H!?iGY4i#G=zt&psr|}d zaUaDO{Vq4W;w5Qfz8=0tlr+`xqcxG|V;8$%WT5ZO;LZQfpg%*Gefsgg%&F0NKid@DH&g2gmt za$M7{^d44MH6%^W_mj4vUGvSN$)4$*!I_SVi)e_?@2D<3s2&WKbf@GJNp-N z5icB%5j^ECW!4UIrPQOz5zBNM)@B;6A_c%|i*;7UKuV1cm)gOc1aeiaTKb6rdpC45 zuO>qvATKo|jW^(?0xoP=|N;$NL!JM42_9C9gItkE#X;leZf7jQ=c~8$|S% z?P6sC>1*q^K1UFOQMJXN3t^DfNy{ga4A+gi!`1IH$(b#@bQ3KYV-?As`5>s)Oml6l zUw2ed6t%Z)phAMW4=x6Nj$4$+yHFucc^plwZ(sbsCs-87`cHdtCD_)@4g$0bbOlBY ztYQV{;P>qRlx}da%#u&F@xwknxZDNi6|R4}-*@r8&OQr|6^kGQ<&Q!|Mqr6m$Ilq5 zp6-7xH6F>UbmWr)!XP-PgFB?;<0+a(HC=$5717xK{bgmL{+D_dYdo_m6EPUHRihYF z&6+Xv2AaYHx;4=jM)cvv0)eJtHeM6`Nx+)^>oRHED;46>{Qbiw1O9gif`?nS~87yAziYJZj-aF?`_6= zeI=}I+FOYF)5pEOmH_4If9vckPL{c{@VeyHY-XC<`%~hqfJ3DMSSEiT(SYfiBOC@i zwBRlR(brrFuyE2b!AQg`nYNSFsinFUD0T+qj8wfFMGwWaqaT7#;U|C#gqJN0bzt(t zNNN1*9D}OUh0zu&r9?81FGi=aT)9s0YRy1v=Qr9*$61owA1L72)?09EK-B6t4s3T4 zo9hjc)aOU*RCMNKgb7(jP5&WEvRws;gf=6It^Fx@S03lUwtN_|wK+Kf{S=S-W$6jN z8E#kfLUIGfw+dSWWJ|g2Dhn;I`wW?iBnXk4Esqwrl2Tvreu@#sC|ECT7t59`nOG=) zl>-V0mEYOG9tgdP1G3I4P!AkX=9ut#LH%<4hcn-Uiy@OcURQ6k%)P947-Hk_bm zGdcCsAw_gxchG{^mYdS{?ROa?Me5ZWmm4cnp%rN$rPe2NB*x%4SwBHyt>6>#<3Fx? z;o;mKF>_SVL^n?Vbvz)>XBr;Qf?=60LlUncNwx-#({bnTQGvdm2F|Z@l~5+YFv!*gb14({ z(S*iNba45jdAexCCy@?WPW#SCg6KQr$Wi?>RfS%0>MXBbV}?Mflh};j2i$GR(Q-G0 z{;!hjvCEXhcK=+59SB22{G$K1L*hKOw)#BH=aD>!o}LLrnJ3qW?Rr*MEqNhds3Xbx z3g1F|@VF9tycp>Q=hCmABQW$D?Jn$KY_h+VcRJ1E!;jADJ3e+q1uV+{;W;sIbaEMl zO&wci)3+xgO`X(w*h81y4Kg4v^f{LfxZBqe>ki}dQ2IE60_`c6>y~N0g~Ohq4$BOv zOJkP#5m^+3SO5CuIdwuWkQlkeSS-eo(&9dx3U7XNx}7ods|yzh_zSP_CFMEGoiK$X$vk4aqTntCG zjZIst=^|*`$)sEEo8XRI z+Fq19_iD(b(OOodaZXLjITj7o>M>Z~ORahX%lWFh(hC6~S&~qu($s@?dqHRs=<~{$ zmxBr)nxBPLvKPmE0I)#^hMG z!1wW2auskg-b2b*pXIvTL_FZlIdwmp@#&y^c<>p|o0iubm0|34PyU%|L$x1C z3Z%xU+dF>R-WL%;K!5m5J!LAaK?aT!4HzUdi1`Z^9g_bIQZ)IyH+rHlzjHC3%~|i| z>1VTZ6pF>!?BOajC_)<)Bq+!e z_jA{eVAkTGD>y083LRR_5A`LW82$3)H6jCj_(#7DqZX0+m+K;AboQl-0q8a{WFP!q z_X*T0}&^U35O)LBP|5bKTvXo93O2p z3$h1U_h0SY3@7zUwYDFZXo%=3;t;15tiM^-e-CO1G`l1>kBr&6xtZOzkc5se==<9Cs@Ji9wq<*R)gO--<4gYu)+#+hq#t(!m=d|giqy1Y2h3|U37IV zk^U9y`UM-;6!_+Gi4GcbsiX{VQ2-c?7{Nf;ERY*c3(L+H+iZXnU?AZARTFIrO_fwG zQ}e{bd^t!c6# z$TuP05Ka)HmBzj5O1)$@sbtrx{N-o5m;u^E@?T;U76^)mQQQDs4ZSCiL1 z(*4xXlPD8CTfjA_`z_i1_xM(zs>cfmFbrgc6mOf)BIovn+tY8jql;HU&-4xClnH-d zI3VqeRf2H3GM@F@ep_BaJx$R zC6O7#A^5Ty#EA+{sF|Zlki+G6pRJy3meLp+q?+D;N&z4&M{{)sw$jnRDEc5`y97O!X%@g zc-@N^Ed%B;qu~ip$nDD}C-R*P!`4@V8lc|R@8N&>aNq2F`;S!=ml%*^UZKk!Jvm=);sTxAtgu|dlTc@ zPzZrg^&AJnV$B4YFFu|5dIk6Ok%5{CLw#NL()KUFwkt_-fzjPVStZ{rbu-{#oM_5rsOEmD1uL3iq z$P0}xEiUJmdL^tnxk8s}ScRE3e#K{< zK9Rkxauajw>=D=SCRw;R6&&%ft*!C{1pP6-(&IzBnm|S zi;tf%##*YA6NW}u@?iMU{!DIOA8A(bQbQx75JjW`o#KlkSQ;*a>XliRe6$`IznWyN zW|lJGUl*yH;gk}1i8RN@S1@AgvR1u=YCN+2I_tGSqioXmSs&7t1SrrdudGG6tgpYy zGR`Kk>N{As(-~^eZshgw{Dh6ny;%EA_>sF;UVRZ)@YB6@DGD0Qwy&S)e-R-XxN+eP z>el!)88tZk%lF86NuS095vqP2@fmBH;E|>^%khTm)J@O180_llFJ$olNC!Cqdb8)0*ZuKL1pwg z`Fe(t7y8&L2GxfkqNqci`^wCKRG+**JeRY&oXy`ggtkiUi`6!tW=h8#Tr2gGL|hUJ zMUjlkG%%9LAAY^aq=~4unW4$(Wd~1;&}kpA_mOCz1T%+CqWg1LMf_zHI;llr#yqT(9ZN0mrEReKxaPHeI}>$zqE*a%rh%Ed8av{ zp?x;aNq`qLl|Uq@X$qGb0~nF1rKo zx8g@dCwsBt)YkFAWJ6{x7eh`=euan|`jrlpOxCm&f;wT&AHp&8(XSpn3p#0=XX58wXo!fm6GSc?{L=_QJdIdJmibax2i?`pCdQz@LhGzc(?S z-O%97$UqbsHHPoyql%^k2u7mYgyD^-G|+Kwjy^va^hC)`Mv(<&?|pLuB7FE=%A#TAcxbszQ4y{VjMauOAmDhJ3Gdu0 zpOMv!&q=28#xv66Y^60$pT*1BVOC7nTV-;lt+?nwS;s6~?Vw)=n@_2PkH=U(EZ`|R zTE9w!`|ujOT(X{2H*xo5H^rqh$O>(2eDLtpq`@3i$2B?F(g+{rAEYfv2i!AWPIS|L z9=IQ8k;WY`rSYW%lY<~0h76S-bl|Avm?@1p)NQYLbiLDI;oNX*qw?_m(-QY`3;&Pu zLrb`q<=YKaRt!!%QAOQXWRE5Y&QsD%afW;3qv)Xu?jK`sZjs_mEm6p97}f0g80Xqa zy;h=ZW5>*ZoJdysD>O?P)p5TA693_VHg&ZMWzsX}Y3l;ORBimWwjo;)r2R}bX zb1UWf?@5~pL)+h9?fVy~;MPjC^Xwzwh`PDv4v-hG)nJMcdu@uARNDwqvm3!(2t0ld zo3qQ9&ab~EViTfM~5dVS?Of03C(0Wl$M10t5&H9R9-KgI4-2? zm^OnB=Y#xrW7nR=w(eP7EtN_-LC`^AsY2q;_sW&?daDl0EH^o1rE&epkQF84tHk%8 zp0}_<@(|f*tTf39PY}F_#capcYF?Lk*%;}O%4CRU z$r__EhIs2XeaKz8JNO(ex|6AVl#BS$T!ai9pC`oSxH5;lv-*GzDVYs2m9<1XR%Oz* zBl<5TJWJ_Fz^1n`cyLg9u@N9~Tznx0OXNTq4 zV-|Mj`TsCeSl?`PG^@uSC~k3*GY}#23CF4${lDKn0!k=RpH9j?`6|-4H-jU9)J{ER z8F}=38jXZ>RzqC=Fjt+_!41!h-)!dqfB=G2Dv9zOk&=z{&+4M$QXn_tgRUcrEL$S=cc~T)1$gD@@eHL-v#RaRYQOm z%0aEr44!-eC>R8l6TAWFD}Lv|rP#yEi!DbVQE{iv1?9x4^E9O2E!y5%C{Wk@aZ@m_ zp?iFbX&!3n@UK``?lZPLRqzF*-D$c|m-P>X`TX|p8zvi0>&JxdPJHNrLy{HYGySYO z6Nsdsf3j4-gM<7F2=!OcH#hq~8hoK-V2c>gx+bvbOy45o8x)fLfW2=N zMgy__)a$7lX--r9k5VrQVaR}%T0LzlxTnb$E`SBnpRe69EU&ZAHzq**LCU1m%`@D_ zdaA`wR52imX>4^Y9nvdvXFzgx6?yYK?15ma`KOIPjxE{mGmfuv%O??RUs;|SILtw2 z*oE83NC-;{=q5FrDBl5wCFgp%T~w~sHQ13S9J&b2hjoF9o*NFI$K-G8 z`zIM;2XN8lw}2g^T7>#ihmze_#n?WIN2Ci7nH4 z3rNgx{;N1g&@;~lG}4dg#X*4YaNSi9nKuvgZ!r@j- zvZ%AE0(?IewMA7<;y|FZ;DK^rVg^H?DjG@f_X%-g8u+8ASB&E};CAyW!ZxpZ_tkGX zz7ugWHCvPwmjjfx>KA_&H7q55B+@`FdmP_EYnHV`<*r7ph8do?&(ML*7Tp>X$!8PX z66y)fJ!|kJA-wsmwQ)ifw=a(s*D5PVu3!3*0t>1PATyP%Hl5i^zgl0_t16)uK~U*n z!#G6i&0CR@r_CJJiSBbdEQsgU%VilICyZK7lI-^8@x>1#woIcpsCiw+9#KAw9t zzR2FZekCl1ZJ{j`XWJ$OaLj+IB!I0oCy|X5Syw6uu3p^&XM!EDE?=ueyM@8RDDiAm z^dnGmxR+8dgquIO0;^+G9=UtopDVqTsMgYjVS?Nvp@;^apJ&6!a?{`rTare)YKmc6 z?DKhvw$Eo`9D!Y@j_;rdIdO1n`^x&YWbg4by91H2A?}R+&2B$HtK!i3huPX%aaOH( ze4=srD)vw2^Y&)MR3Lx>2ngpP59muB54Cg#EN!bGJ{`$@DRF06@wiHeN3yoH53eSFNa0q5miw~h%r4t)Z5!C54)ae(Rsf<~x4V}IyeAObNlY$JQxs%4^hP69Fxc8k+ zfl9t%;!=_}UFa%*WjZmr@@UcE*EqwqLpc+)v@Qi~_rG(2*5I~}o{*0bNsKb|`gBQk z^9PnNcZtlmpR&TdoWw~{>_@v3jC16NevKK7e3lax3=NESlh=5fOi~t?ip365Q9_Aod zYZED&!OlH6G{I@w{qOAyDGXETu~ewIkY7eMWJnG6KqaEtl8L=ko9$$4;T`1nCoSAE z;2%%O{c3{~2iUL_4(}v)>Po)pv!HW64Rg%h&l%Q<(Q&&lAzU4qX^n<XqCAtt2q7 z(4`;uY1C|PJHxHhm1X4-5PC2xsE^!WZ>zS0h=;7TA~;Q%F5_z+S`zz^ zo1>y)LZMGy-{=h~;S%niYq2MvMNE*-O1(Gf?Xh9v(9@RFYa~}4DZz)Q{GNn2q3O}l z0#;BEvLnxcpXWnnFsInRU+;+TfYiX7yW9vX6M;Vs3pJF4w3YwAFewG>hKXxcy*Iq6 zWJ(gA?=A%>-@`uy8K;J86rxFwu)xeA=ze1$-=M6$hvz5~f(onGtRIdL2nZjy1d z&aE0P>3U^`wGo`gU@%Uml@e9F4@KU~_}MdxmkAv7vFy|SyN*X^sE-p1wR*00=$vHf zADf#TtNDy}4$Hpew!c)BnuzE@E5EuvYBTi*hr8sNy%S2?kf!YO#u%TDr1jM` zX6~hFik(hM^(GQOt=MzC=FjXUF3xskl!`?dn<8*l<7g<{b*-5*Ym6LQUEsP{JsQnwiO}oaqH_QR4j#?d%X3h{YgOAOyi-7HSgM634s+@P@VpjRduk$7f z-}_O4%Ka_Sb|j;d+vq3yGh9~y9B-h9&I~4RvVH9T3`Y*|SoiRu!vF^U%bz|XPVGCR)a(YZKs)o<3QP%cK z@(5$5*uyS{)MiTSM^ga=PU4Kl_(HvtL#HdI8Uu`xDkb|Jz01-YtLBfA@j}ygjZl)56kvGDg%EOv6fV2Dh zK(t24eJRK8P*ictT45arqOVB2G=0;iYpOFvMZ)-QsAG=~nx^mtU4tmJzSfDshJGHW zPt2a*O;shpw<}h{z1Jp0bpV?fOb>e0aYuS2uT3l72Dm!%_UD<};8TEb6^Ed;-|61Q zK$rN_Rb;jD^ZoYVzxwlEG${lUBuSb-A69g6E_NiY0szfwi+ zTXCb%!_bg5v8+YY8iZPVbQ%5B;HVlI`=p;#2^hhNEa7m!)3sKxKZg@8y|&8m)>H6; zwg$I!%3_rdT<@lddR+rGkG8xV4oi7xa50!ljk8J|2oKrm!Wlp`xKnWK<)cxUA?&(U zMFB=tW!2T-(7i0o#_ZOA0<~dto(_C^&CH{z!n*4cK)dqbZM$dCxWmW-#FAv{*Y zzNrqoX?&qh+avk-+?&$cLjS6Km;EOj{7+FK+Q@uf!L4SA?Z>-d`@eHW%=DHS^wREV z9iE>Hfe=m%= z3x5}$+8F~k5*5zvZ@SJxtGC^thrZ(xz zRMK9%d?aHtfTJ)xju^P;fTNQl$N7Xew2%o&L#M8%G_+xZG4}d{o0?x6+n_Bmr!e1=X{>lS*}~<5r3gZ$@^d`J?sLOIK#?OA-og%%nTdj2mTUJAL&lvtq0TAX4pi zyx-y~8j=x>n*$G|QY=1dRK7u$ity?p9ut*{{iuZ*ANoW<>m)Yh{DibU`VR-jw|MK{ z!*PGN7)59PmxRaxVQ)7nub(hcD#wrel>B*$)lZ4%JhQ3sXZ7!H&+wxH7tA&eEo?&X z+m(y4#a`EgZm*ho(~xiD>rJM3jz;7Nfv;vn2uS->dfjTm(z z*rpG%Y>F&*i>v-4f%PsAtS=|gc8mL`&H`v3!k>no<_~sntY|ayU2KE8l)ujE$2~{0 z#cbA4`XJmN|F@k&5`yKwEp9#<rks8_`z zN3eIdCvF6i)EZ4VZB{(_#!oSDZSv1~`}&K$NDaRLr8!B(=D3)xQ8ye0(3>MIt`6we zW!3<2drrPjB<+jkD6NNo@XRN-M15Lv*4s;*v;NJv zUaj#c4CiAAo1JGy(=IH8O|Ii1FUQy*DJ3-RdNtNkrMRtSlqtl3I=FmLJ4Sg3FUl%6VgDY2AI|YPuZ6) z`^teBIXlb)gD!jxh;avuPuq@L(l%n#0SONmS~GP`E#{$WbzLhRO>TW8PbsbP+-jX75cSqYMGj z3Ll)=&XWv&VszzKB&iUkP%w!WXL5w#bVd`W991r7gG#p+X7&-t7-B)=`U5rw;jBOX z?qmN0n2Nz=9YlKxB)`@NX_U6xfrLN5)XY_|Of^3CQ`$2))hU;maCqpcS;PsDjTn$F zK$&f~OW7s8)O%P*9OiI5R9!qjt}xo^-G#pV&7VDok65UQ4^7^2k*#)>AsC04@sr~RHm=Ap|WpLyt< zFy%|wU{;;zWr5b$iFX73;`G#`hsH6(P4yj8DmHn9GY5?~6I&g(vwSo-Or5RvpLsdG zelJKV`LW)?sWWDjrYtb()iFOaUWLOPo73-DDbxoM@@Z`xhB z(`CViUu2>EHvI;2+D~kRGx^pNvj$)AY4&G0*hv#fTAOGoHCjMy_OVuO5rj1KfRkhv z5WHg1zT3K+kM}_#zH6H-uk%SZ+qc$srLT~>BqOGhUq!%`G}oO-#l`%3-YVelXkouu zvd8Mn5~L~Ss-U!p)j`yJVc9ue42dPNd-cDaduz;S#XYo`Vp2IvA9~aEqE7z57}EQ} z;na{LnUv%~jsN8WQ4T;@B~Q~2-TREQu9FJgB0zaIS*Wd?rQ6@?WpUn7tVwnD+*KRI zz^$3vmNGGoYH8M8B+iVLY1V&Jc<3`=;rqqp2~VSorMEjh?`LK}qL16eDhFteVceud zT1nXqGseWavdCD23}f#8Zj+POqv05}xI`=HXS{K41hyh*m7c9^9~ewPxf$zJpj`D> zVg-ziK|(Tq;A~|mSd(a-^h}bS>uZJIiVi>p)F8@#mqcH}%&!=0A4T-I+w40RD_*ie zmkgUne6J>a#=@L%hs3TnK;eriBA7k?&2wD)J#IKX?~m<6PRGe~!_w?V`Q<-&Qjd`QN#-BbW;`M+ zLEt!UgEf&V4z)p1k>dlaqRDzrIQ)A=abi*Ur^%9!vZ4J0lSNjG zf$ww4%%T5>)pXbb8^jgC+QiLEi=nzoA004D1Vwa9<^eIvlqs$~TjP z2p`m(t^U`a@I-DS1pP;sfn_354oxg{tA=hr+@hT0vyfu15@%2jqoq+!SJ?I6&Y{Wt zzUi=&&^REg|9$Yy)g>od&<7ehtXjoA$^K5pU-mbX3|HRmDz1jg=^y)V2q#py``0(4 ztzzk+-CXIzu)YYV&`u*XE3=PtZsfp?u%!s!yI*QwcB9T8YxIh9WaDP*NMDXkW)HR( zeZ>wTb1&y=_5O7Qk%f0mBjQx>8j31~uA48`lX?!sNF1iR15kI)2ar<)b?tT%y^3aF zrZh7Vs~phlN1O#Ge}}0+BzA!$N|`R8rAG%kFwi4rfBKvx+&~h=2d)Qt?gTgGhBuoZ zq8sVZrRoUZ>D=|83Z*r!x%&^fOAfYWwJlZ#e1c}RT#v=@QSf#@wRZNFDE&1FNRmVu zd>bI4D(bCu7ah+qE@cZ0LGzeY~x& zR45>k$S*wAVAkTZ(`SyIR4 zKM=cQ+M_BkdK?KYJKc&ww|~oA?tr|5*dJPTX9Q>-k7y+Ly%9s+Z})utKOJ(gHpr=y zGOy?7@+tHvAmRf6bpVLw+YILRb368A5)+nVY21*bAx`=t*i|Jsn_?tX$elS(sTgA> zesw6ol%Y@qmw_(kbzVhM=4z4WNyBDYjQAdR_@GhL$E&Vp3{Aegr;Y`_`CI!nvqQ}7 zFm{KO%vSYUkUkQ%q$G;xIgFFz;WwUz|G@?amd-{2^pI&(J5OP6JXUp@Alot}$2fh%DSuIBc8X(( zYkmxnQ&m>~Ckq%s1z`J+5ooJn3=Ih$T>_O#8w0#5y1b)S(LLLs5OV6%a@p=U@5Sd9 zNpXj8g6E~7EaO8zxXdiIfVX~yRglcDMw>qjh= zOYjYiKUL_5Qps#IuqWDdDK>`B1zSp>T9+qKQ@9#Y|JR4LN3y7o<8g0{>3T@N=%xxS zx|V8WQ+Szqo;K%J-h*Jr@LG*|$#|?eEod zWoPXt3O?bktCEf$RW`COq`3kurfbdb|5R(I#3_5-Pf?frWkC5zsJgZ6tRO$ZGAZ5> z@53#N^oX%{^TVRB`ZPs?H4BZv$8SFMSj~MBZm02a?p8~3=GQk~YI{DQA`vzZ4;5hJ zyZiu5VV+?{h{-#hfzwJpDQ!5TB1?n~zbONgVEr&#FbStC8Ws2DG}u3l6n)3zhArPW z7yzEar8OL;Gsvm3Ytgm??(;48q$KOqt)SGa)cyRiJ@3InVcup$+DtYHJO3{j*j+fv ziZVOnZtEyKTO3*z8)AgEs~9QxRiIr6fC20eXY@U%i}EJr(x@@a)8g@{0rvfk-4cKE zsOTlplZAN#k1m8~7s{`FIs3|aLu3kcbY=7_BqTEeb|#mOJftFi?q&{_`~IpgYAunC zijZrweWf_KEvEyTJRq7XS4Ui?Gv`N%w%%owUE>zbz$u8@cEOzQ?Y zjks7ZQI-sqZ&bOfvMOdFztBL)bDehODO?w<3`*&teZfW(*mv$5Wblf*s$o)nuKEdV z_yznP*zR;723s7v=ekwEtn3FDtxSqCO(YsU*$-i91^wxT=NN|O;2h-`I6NF-@iN3! znq+ZSnNha~kF#h4H!0|;#1xQ*G1|%?t7nF2HEw$+>X&1bwYs8Qx)>BQ3%VDr>omnd zhNG2mww|yQS+U6_vnC!~_`T|bmeStW?DAKegrHEfsr9uob3u$oS^tT3LDFvH7!$5% z3uD}EO@evlJ&fVJtgODiflr;uW9b3B+$7n-8Oq{uIy&jMnFlqF8fAaZ?UrX(33yIq zBU!Y~*Rni%A5yQjpM63#{!wnI63``z>i^oZ_$F3GCVp{Y14gh#YGZ!${ks(yrHL>B zxa3U!X^4?evQgq6B(;VKln<%kkNVGhIWsm%KTiap&P)s%yr_X09ij!M+N=>+?%v2_ zmuk1w&zG)if#@g3<4xE1LTmd(t&uvE(di8XK$v`>$i#>}76Y|6igCSIi#=OUKeCgy zrp1m2zVJY-HlU5gG1hy2BhnWucWoCqmVOv%`23?{Hb@NT%UqKu?`La#Oa+sDjuISq zNBzruF-Yy4aOlf~roWCJP^WQJMs}8&(Sl&#ZTFx3a{0o-Yi1mY?{=JT>9*McW0fS& z1Z59cgqv`;*4uVE*t*MGkQSBa1<)p7h*Jvr<|LPd8I}thKiEYJXFv2c^d(o5SLo=X zoyX?+-F$1>p1msy`Cz+1~ERO zPRYc6pPvN^(KdKR=p}r!Uh@}%$M&u67nar0#miiCl=)gBALby4O(;$+vr1Ka@*lWl zzao`{h1HEea_BtMD@Mnz9Ihy07tjgie@ITgLP6I8Qd40}K%o(f9Chi9cFkZ6eEsB=1&?lYmW!$8M=JCr`zK=ic3G%<3>!}vq9b8aG(qd}f|dfdAuWb~8YRZO4F z%;|*Bq#Jd>o#IApa!)R;_ruD9Q>pS@4v0$A20G#N-O2sCIoH2#-f1ob^M6pWHS6^X zW1(4(UXX|&w2<9TRa(_<+typd1!~1a!#e>bbu0UJcaR#dmzt)C`-DgW8{@~=O)qE6 z&OqGwjUSEQ6iN?CRG#LlE9s0lVps9bEApa!C?Q)DEnebH3M-$#m}s1pn`!qc0u~te zc`fod25b(0)+PBkHPAMqECP=tyyuzSGF6yu{(CF(L;eUSZf;>0m2P&cYB_ZgyVTY~ z&4t$f5G$?fa}cYZ$rv2uSWMJ8Xxs`tH7`zA8bW1+|Zi3K}?^?Opg z%cmMAO<;6(>u{=0)FrFL%)p06)N1l5M7nt^ z;hnF1C;SG2`kI0(vh7Xq?@xH@NlZ%QAywF|9Ns#?q6EJK{ER}H6W?{*{#IDaB{IUz zzTy9=0en8aZB3w|@Ku(MhYVR~NPFEGmtb~!ca;WwMwwUKMev@`lh3N}Y~Dxx!P1d4 z3<=2dpRQ#~MbtxI;eI|Ly4d+eUv%Z(wb!#D;3=3ooQEXvLIeHvIvM+`mHh1}E=6tL zmdBd&u5*-ia%tp<=4se4V0sB?2ef<_?^~+Cq}SdXFv+DiupKgP6MHaX95#^G3omV4ep&(jRL(580@U@&#E4I*nzVG}#3KgX57#kk| zXe>r(ZEvn}Zl$eN3n$5ZN^m>=!cZF6@nBbwEflOlc|b?uTog(FYspf>f3&2K5Fjg# z(vw+{mFpb>zrH8!^e-o!#jsxW8aC=|j44l`Zr8yXpKM+8Nu`zJ?y}Qj_@6-Qi;BPZ z4-?7+oyU|ZD(l_)?|&BTw#yv%&KZWTXR|lZZRPUFLwp!^SYNI*aj->*?O*{-6Y4T61MQf6x-Md&25 z>Bp>DN;Pv+jvY{|8oH}rNA({q1SLWsw#%u9bnLwOH6Ct_ZwKvHD`0)`y+VsUu5nJ> zihDQ+l;Rx^qblvTZBQcnP#__(Fc=mX6s8k~@B@R>Ql$;^Z|9HAp>Bul;}BHAe^xRK zi4?AuSvd~XvY)_lSuDvV|D6P z&Xk2-xT7W6D!pIG-%hVMSP#qc6Ciyra1~}~`(JYVR=IT}ebC#GH<)AE@;qYxUrQEX z6R9ME>#?8kyC%Z^+UwT{r?}e2JR1F&);qw2dKX%-U3UQt1<`9Z3Qm9tMJ=6g%J?;uK-j*cmk5fi#}sgdsGM5@ZrH?HsVumw7pJ`C zZ^3hU`1Z(N0#D#;+)SRl+}Ck0)ob*;7ZeV!wmdR7>9I7MBE?$6^LP6t;p<#AtVQU6 z$V@MbYLP$9;q2m6)toNqaMK8^jm^d|6sbFoUo|%Q1uVFI;e0w)_vrNz7<+T;06raJ zTMrQZO(0ftQ-;XO`1IC4RPkx-^iPD>h}+NNwf*@FAKMF_(!AMARh6GqO0Lx(j0f(m zTb4LjqYjf)w(_m28m{JB2e@zv!vIEJ0wm=$csYOCkyGn~@Bzg9%(rjPgBo$DNW6J9 ztvbodW{1Y6RV%;;*B`S|)Z6yGXvt~$zSh&zKxV$BqZ_z03rrF2#By{ zyb^vT5(Xe&sIAa1E9AG-wB6%hba0RVN?52N?s=yC#SvFd7iS57r+MIb=vqF^2wWU3 zGGR*UP4^IMuj+rDKZ@tgmYl#c<9;E$9*2-@K9oypsJZE--gA?*tyvN+Rvo6tI(BHc zrL+$BMoa>X6^_$McXAB54A1miA7Qoer}*l(59;7b7!;uay}_q6hx1VZG5#=2m$z{< z|BzI;P(XF~Ss(FL*5Vdw*$V&=eUAxPVGC6Qr zhddffNIyMV@@wR{2()~%fox>P*E-oY0TmHJJ(G;OT+_aY@N$@Bp?O~-LVa8b9xv7H zYOia2FN_4-{y=Tt?6iNw?|HsaK!E|S z$-(&5`cb`sU{9T+0uEI|pvn=YNieE8(z8fv`A`;Z*wiS?R5!JbBI-&pT={IvMbGl< znI~6L)=gz)42@aHwK{8SD#AeQVjWTRW*aOZ@8Z9Rau+Ue3~Nj!2YAV7skOfN4!lp~ zMX5`XkA{Uxko9qptO04F`O|Sq)B_R0YpmsPI7de_b}?YNE8am8)Jd41 z=9%GDn7ql|d-v?w-E&M|wZy-bM{%zPeTttOo3_mU?9Z{d0~`E)7U3oKk=ptA`_j~( zS4P>vN+U~94&drFZF;Nb$S#$?w=BWG+@g=%IS4QSN;~@emrnS5NLPOd%IR9^eMM}L=VTMpVTC4kcn?u(q{Hc$!JZcdIl9w9|?Szs6OsEekFj1tnQR2 zIbNqz)XdJuGgxM{$Y-6+(D5*&7U)8|8M?fYt&J!5r^WFU9@tgqJv!QFz^z6B8-WR> z4o|h>dFE>lTWUuH1`5<*&O`~j?6t?nS!&TBFPH9xovneIk13@xHcWp9YC@q$?=b0Tt51h9C{`rHR)Qh zRXbEYci@3ByK2JK5){maNJ(_3!kA2ue`dXAB@(z z#7NwZFt*(#rzNyCBZm4ND^#!XB-z;*|sAg$V36&l4V8p^mi89*RCX0>7eJ z`iqh?u(f8eCaz43QwYo0ZnM7I-ZAudwxt96lgujnFOG|Df5q8m%USX2i77{2$wzJd z15SzZ2C`w;&DB;Z%VEiB9LPM{4G#YUPRCf)ID)vRYox&s*AW6_{KkKeOM>Pv_l;vt z`?i4Ifa5uwb(&6oIva)o?`rm=%h4+seJ1I*wEFTTuL4zwmE;r{k~BEzvKEfF>GJxh zInU=V)bt4tXz=k}>SfF|zPJDazf+d8E>n`;d)N$?35;wT-#Pge1T+iOLcDS1bWZ|sHr}d`Bw;Fl=wwEo>!S> zhL`cDs#Z98wGqa|vGn@7#gQ^YtgKS3Q~Z^HX~mH-Nago(l;F8+UV9<)q$rPWYp)soR`%3@mjMHR8o9F1_>wK2dqUugVqU1jsEnl{&s~<7SeHnRLy;Y_bLDG9W*798BkmQ0ss0iB%5;zygd$tUdHhM z-2U!mzz0GjuPnD>CY)Kz(vL<#u^~z#}RgXxWKJgm~RSdINJNe6oJiv?LvDZK2OL z|0GW{?o3Xm(B|>djs;$_6-Tzw78fy?UIxEaa7v278TXV8ahW4hA>RQ09fcD_c{ln& zcP~k&0QBs*hC0ss9#nkTFgvs~1Uc%PJ>?wo6lFD2Liw31>m=VS+kcpjtsYYvv3d0P zne~s85aCgU+x?hnykmXsIYqs496q#N02R{6jP?A(Z3de&_GV0ag4nM3mM>-n=E~f> zFzaHDt7u*Tx>2x{S;M?hzqRA@F;FKYkR9~u>CV8m@9W1+uOJ=9Qe;v7sR(tKZe!WZ zEQRGW|H!XnsMerYPFSb7n*H=E7d&Ok(DNuk%Ru`&BpbYZ0E!c90F@w3u0>mWljcEY4iv(7A8atZVfQ_E*`=Ey;?eZkjg ze%AU#`&&TQwAQC|)C5FQGaTht47EB~E^4DHTG8)H?iE(oTnAO{Hn3Cg@yU9YkjxE> z<{Fs}Ai?nLc)qu4Jcs$7@{813(8PB)P#a#ruAwSshXC}$n`KfmJC2T8l09n$491wS z%2HO-?3gCpzI1M|MfV#&b!I+?JgZ8Q*Ya1~s(y~TD~!uApOhA^TC9?ThYFHC23RP) zH!VnshQyVd@5Mnt;{20{tr({_kS2v-`I8Rkn4y_gsDt)a$H1mt#AUwv$inoB7A3hQ zu}>*rBvh}???=tBg`xemW^MM_vH&fkb+4Q8fVy#5oF9-*3eP>&HZM3+ul)y zN}`{(??us>DUhubc>J&^TnOkdp{v|HpMap0R`nn`-m34$gZuJLzbL8$Hi=vgpEv~% z-(6~etd_n0Z+vxwqPkUqTqiQ(6ERdy@h?Fj40S$KXyY2IWDD`GtTxB01yiB=|4(wm zKNd>o(o<<9fGq+r#dewZtk1#fjGEhob8r8!qUv$H0{VtUzKwx5-I;x7RaETHY|NAr zHAVB-qTy5+5uPYeYl|+ak`Z2S1KgtcvXo>ClyT!FMH#JBf8)0mzl7mt)O^WD^L~>V z+9YWF%s)%0GtD|Vf^d_=0et(3JN-yUnu9o1?2n>iU*7Z&`~$pmCzn-ufybB#1^R-( z>|27Q3V#5ZZiR}i!Ln5)wND40=t%8M8HlUWeH$>foeV#bAMeM~;X<$_aABmaY4V_0 zi41N1gGg2cMlc`Kt)lj&7+#2}M5c}}GLEafuiu<)Bizn`DMVYEuI!|zM_2fIR$Q(l zF52`Kbh3%4p{TAy!$fc@;krmCX!dW3aE5hfUj%sDaSgd+*q?ih)^K7u6Wu@iK(;o# ztCpUuL%EGacuij;4jP|HYycnMpmc4sND3u48GqJnDsr~UOyys!7=~(t%EZ(i&Uy@S$E{ma5-`wmvyE808AqcPr+K5-%JfzxWl;RqEYT{-r$c%C z)Xb+|FLZ__T@``mPk>t^nvaY5v<$6S+bh)psY#9c^tktO_u}t(O$Sz&kWq>*ZRS2l zP}yjglE%5XBX#(6RhMZ-nt9aC`Rw!L4q|pDXWC z)?rUV-d4$gyiLD7LRk4V>-mq(7t-o}E8Z)hkcJXpleh}&j;==;OhF3?GQw#|jI5_z zQBr^S5?AiFI-U4 z@lkauTDnl$uK3_1XOBtF@`sHyv)LZKKa&N&m(;S8m zXn5!%@uN-PaOhBIbC%p361sw+e++BUGRX=6*B`oZDe;}bgV$v)tzd=R5sh`wYl7KS z*>>9wx4C>mS5`To@&r9`#K!tk|L^k*{R43O*ZICip*tV7Vo<6w%<$X8S=3HSuM(p{ z)tR8x3O#gv&gGON%Za+h)}&H-;O}Jgzi*mBIr?ohadD5ydg*kZ!3>MU9B>n?w`nbZ zS#~8jIwAgu13jy_K)POjnMPJ64>dNu!>qngw)!LPV5B)l;WY_nzooizT5*2)CAeX; zhS;>He+Hdssto{ZBfDndfT(WKCU&j0q5Freo^fcU`Q&iIA&$xzY;fgXD?koBoMFqQzFc;^B%5FLKVXAw(!A6x zAGAHUv$`n6KX@*^mK%-jI)KjViJyV*Qikg|+#6q)5NB|(n)Vvh@DSzEji`dQ@87w- z!)60KF^T*z`^|0$LH49@Kn;6iYI~WL?6A6?4-%k@A@Dvum-$`Mb3e<<ALQapg#MfQgid;>%VqV16MZD#-T-LYL#>?R zJU{*2bUen^P^~6oA&y1V$0)D&;`uoLbLX%6<-Zdy`Fq0e2C1$b?@`XcL|7cMH(<$U zC5k_>qW43R8e~^KK?|kc8)oah?(<{L?((^6lUB?A?aGtNA#t0?Ocu<)WGMNw@ab?a zTbDNcK#xMVkU>8_bn;Ai-8N@-^G3HAIR{3m1v;j2hjNdw5Dzs+-TIVcej0E=)pAa7 zG5KV@^`WNpC!Din21V_Z1kr-?tiadKZ-Pcs(ki*vLman01S#;}BDz==v{`ye%JqWe zRjKN2`cZJ{goV7v^DBDUtwbVVHM9B zDTOmOx?JubJsc^eGLOzP-Ew}0w4{)r@A%|LweYt{9V>wa90Jc*Wh(8@K2u?qBXC5O zpzP9C-pE0F5nbZ9p&E)WQ*e~I;~FO&oRCYQ7WFpO)5XkcW(prmy3G>r3-_f(<|2Rh z)e}p~0x$0d%g;?N70U(=P!YS2IJB)v_LNr$Y7qD<7!ZOGWAPIv4z)irfktEcV7*g&i zY2`R8DTn`hY36@>X#|x%x2=l6&aBQi92E&!Wp3XL7}q7wZ6kiLJ-X~+(eDs=8(r`k za~x8;AL^22$N!3kT&qn!H))r{y$bZX5WhTGIg5DqUYc;p|g)szO?x z)LI%RIONk0Ye71A`Gfr|DeMswi?dfR*6O}y%44lJkfZ0xI^pb&xeAtc(iKpYvmFx@ zgPu`Mn1%D_JfCV>f_+qiLcfx(f67dY^rb!|Q&Br*cMS~;C_!}W++7skym|N>_QV0C zKpJwM9ZTlN&>`*KY-lhYT7PP^vF#mS&4ZVUN`9lJ^#|F!Ra!RZvvrNL$BHmg?rjQm z^q8a)b%k|oz2(!JTJGg|KR(zfel9DVvvZxWS}h6SqpNh#4B2kyis#Pz|7BVVe`aRw z&Axhzunc71OMz#OaNmbA?GN10)L<0Rbs>}Xol}nln?DjOhu`n}zcFYc8Yo@ZEBt&S zC;EwW99}OQWgi&Cy}XD)mp!bFI)^BdJ7)mRG{!OMqJFp_xWz}U<+dNN2`gEaujO9S zeG;q@bb}V27PWvJszc>`0WzFCV?JtIp2ypu6=+o&Rp|IN{M#dH=mt|eiQC~v z{T|v~2$y5@4EFU@r!0bf$e2bfeF zOErjrX8%^e2POWZq@XM}x-6nL&7k)kzmkGJ=Jb`8cCaDptXa~rnP$~2f8vJEsYh={ zP4q;ah&@uf`2mD=SIzei>OmSu(#v$J*dqvzLcX)29_x7Y8o@(R&?`0TrLevIi0xXH$|XaPhH5q z+Lysnr01?Rndy*ofsz}?{I{SR!&HdN3r7cO&IWX; zWA9uPD}l8{T~||Te-zwKq+`!Hv+EYzTLdwdoK!P`_dFL@j$S#F_X8-ZU8e-oFr04V zKd?ag1xbOJBp%02sOpRDDs(Z67^>s40NrAfq^;ij%y!7SzFgF``&IHtUYTZ1nRtSu zA>APZt0SR5!IRKvCPX~q&LvGC3aTvZE65rxmAO|P(ns(&w?P-WCu@vrK;E8Bb7gfV zUzkCx(!H3TwgylsvKo=~AzY@q`&a$ldHuLL=OYdB?#q8`wkCh)n?F*%K47dPV`;g4 zlBK@cc^1w)25Z%@Xh7G!pU3QK73dNo(-~N_`@1}*W?UF{d~ zS3G&4gCbKQXFNk=!&>6@qCmy6l@J_U?UYIw0pGY%tJ-Emvuo3C%MC~94^PUHJt`mh zhNZl`E|__rgji8m%L`FvZQTd1o2l>~TM+m8?CZPz$=+w-m2f&Z!ndJxW`5VP97T)U zEYOjrxoajw4y9zV@o95^wVN0wt&~wSRU83RJW&LlGgV>bk(TuE4V8CLA--1peTI2s z`NtC37uA zAvOVJ4cI?Z;^>rLL%XKM8#nwgN{p23u+pH?jZ3pA&C9WnYW}ZCI-qqDhthRv)@LD4 zbyyeC$zY5qr$8s^rbLR+fZ)P>KQcIm!>XV}3b^o~M0~m_07VpH00kLJsFSO!Yv8N9 zX60J{@hIp%>+-CI&+MDow>3}5@o}c>SRN4#2ErGJL?o9V1Oafdm3Dl!%=PF>W@p!_ zBbU#W6qcSCqr++h21g`D5FxZ*ONEkDnPIq4yHjR`{?8do*Z`ojS{uxT%lNIV@UB=k zQ75nI-Me7uKl_N_nt1wF3En!F#Pd^%D;0CusA%Tn{LifHI}9w;I1nbETeV99Dc+BP zDD3Fr>WPVHPQlUcbZU^j7V|sa+))z4(p!kZjoZCmK7$O`)bNV>Vy~DYsDIHh$W*4U z4OaEz4_6msS4$3Lmnb3b=>1#b8_?S(tmO`~xVSBr5lRny*~-g}hgaX}sOC8XhvA;r z56;<2*`e~a8NE0G8>2!aK?_8X^2gZZ!J5u$@+Jp8=MM9*x94L0LtjJRUMN!gx>Eg2 z)lS0kZheolINU2`-1!9t3G1`Y1*=fWu-6C@6Akq~wLa1fu)7fET&q-EbPN%84YO;d z?Sq}$g7?s~bH*AuL;9?Dn)&GOAO1d_*dU?#OX99+T7DlShNHJwpb^N}Ugh_;7<;$w zdLUO=U+^;KG3Zi*nFn&hquQ2Xw3ELACqkYu&9$#(LvFN`_L3<90< zJXxI1#|w4^tl#rGu}?0y8}>_vC8JZwf$L^)y%4I?-dw8ua&@Cz*bW}m`#%#Le{w(f zg(VJ){TtJ*w`J`(Gz$H@qwwWAmG7a3bfv)cnv1sE#RSHGsO!z(hRHy=1M=1!5F>GC>FJs**8|eB>Yb+E zll(a9iAT5GWZcs#a`8g2*+^`F`B58@kBC4F>bI1qpPtuW^2G|mZZ6Eq;@c0xI?SqA zicIzjP_p@Wn@2)m6{t-et8b}jLQP~VhN5J$!bbr4!)ZPp?wDslF9r@*mLm73^FIjW z*|#WSvmt`h{O`+h@7mxKTrNr-fNv03^yhnSjl4LeE1iJ8)wfOUGF@S~Up{vpK6$eb z@Av}&>NqTEjmf3wpRJA(w7)#t^U#_07us8cRYQBM}g^kjG!0N$llF{g&)gK z=jLnkG24957GleS93mUQy(A1`JNTMVu#=1%_OdB>{Mq7ofKkq01&CfH( zRy5BnxkH&Or3kWh5t#9E@YAou-Zk~i3`DBYoV3`74Sqq}=wVTbZMd|}DF+&;h2&rB zAt5;Ja+N&k;S@rK*w|jA&rom4KByIXmXN&F^iOmU)DTNuAF)Lf>Wv40hZFHaq1Xq6 zU4V)5zjT)4k?F`b{p>3&p_%;fal{#;#;4sQ0X;qI7;?dsKs8l6P^l2gGXswXYn*}9 zxs%dCP0-Y{Bv-oB)OehMv>~}9*MOmWZs3#up(3;vh#WTD)aF`z748j8`tkGVXMBqJ zSa~p6?THebP0((0Bg9yBz$L9L6CxcrkL1%TePwc%L2VMMZQCL>NVKx>@jokU!DkYk( z=woe22~NKtmC46`<_!B~4JD<&zl?|h#3CEsYh9`==FEt@L(@9izLKwh<^H#1W#Yp^ z?z_of*Y@rU@9e_|@`a1~&43YU=08*T+rJp*SX_z>6#~s3-wf2P#t?=bMo-;#xd~qI zM2#5D{GfHd+?LczNn6yQ8RWFa@7|}``~Yjsmfrfqcm4^a-#GAKyWWCijBJfyq32(! zy0dVNWY%sh84M5WR&2%P|JA`xsVO^}`-j1=MJi4Vpdd->tpL_cmvH!DO z`xn|D7~DFkwU(+37xVf^Sya#vCG#Qsel+qyD}y}n;v4Gg4PlYbHkN5N-9OJPsEwyM z^tSpNm;N%Gl*!AMk3&m2t_K*AGR49N@Vly1?j=m08fb!kqC!nPJ7X@Gx zj7z(j@yFP#Oj6-Q-Zb4x$A?H6Cy{*x(9>g>;LamYWCokV;cs>9K-x{^7mI1S=TA@6 zIAQn*!pfd;ydKU=(!kB*I7bP8PJ49C9RtaJPCGsKjq)^+?@ zoBlr3T?bsCi_2Ui1pa$uLtPN$qDf>Qm+EU)#99cs>HH~(YKn59UeRC3iZPqr58qV` zCrY}@Bkqml)ywnejo=jhmxHq2dsJf~N^{7jI0^ zEx^DZfGYJ(GD(HJP@TM@SsyMD=^Q z+m|!0g`(=3)iD2;`38*$~0GYzrlR9%~c*ihHUs8zUSgCm%`nU@E-= z+O+PP2I`WqAFdRsQE#$N+qj6EQr^?Z=A+4GENzGL5&_7(uzEyN5lI{jM!bvbdBmqY z5RxyNA7By;RX;XRPu7{L@YLif2e%EUMehq&oq^Yij(}yYrH`2VpUWy$>jQ+X$zd)G z+R}h$NTt|8kf&mvkT#~=p(x_j(4zDdm099*%J?PK`CUm#92^^7DU|>LmX<=H@toEN zHeUzb5?+?&VX?K~{< zx2g4HPksypC8#!#*4T{`{+>dC-oEO6bd6^4gl#BF{hEXK6@>xbCcdjmXjZ2oPRtPF zg*|vAF=)W{7kRKUueUshTi%W5fwuVLGjK^u>GOG9BlMDzJzzGse?zySS09@?7wLQw zu^`#f#rd13F{?1EFaMxoqKu@A9vsfcsJs0c{b$FCDTNSpw@tPL2${>q8?Mo?rbSHK z=L^FdVce}@B!7?)foY3$Y?pWXp}Ur0swK@8nZj-rnB(KgH5n_~?}DiWmq#r^ewNMQ z6xUD`##)`>talvsuKU#bc&ahMa~f0DQ*yb?R$c6h08Tm8{FWc*bR<3#PuBHjeN95k zQSSTHWn-=V0i)HbKfx8>2*ra*O^u%Z*UyzbZ!5LAl9eiXA*t+%=hQDfVL$}9vpM)G z0fzfuSd*_m%N&4vBJ1x6^M>D=08kx&ke8RNLL^D&=M z0U(Y5*8G&H)zEOWX)6)mN{WztyuAmUB1w{m7m0`fVGgqIQEMCER5=ch7H>Ogv%Tul zSJYLnUI>2kvC|2K%k{^IyH`TS=>+eAQH4|C+yAvWfZ2nKS__!_x8g zb44cd-_(pL0C@6z=BlypzekLy!F3=O3^#&Q`XMH5u4?|*E$h^rX?|x=K7($aJm-=e*``A5^CKngPu4;6 zzrLfibM(+5FbQ+a*fHcu>o)qoHislZZax+kOr=jaG}TE(dsHm~pOe-C|g7J-wN1%rR|p z!}1xS*}FcAl6f+~<+ZwDbREc828pHqpP*&}M0h*eSc0+F2gZ|9-DM@w(~VyGORFl) zt8+>AQ@ghU+>0%I{oe|ZM?*;%36?ilvcAe#I$*=Pf_FA5-s>UP@X$|*HXu`LTSnEZ zQri{Y>OIi@178IGswnf$gAwoYHnn9?-y9B$GD1IR59+FL+%vb~1%D{Sw3g{c!u`(` z+kc&mtdI81chOKPw`3mg#g5vPa!l{%Oz)0uAPL-F+Im7`Myq~G>Usi28Sl8o$JYCh(@NU&oVNnWFRqa)tJ%WN#H9-mY|)s67I!{z*TPv43- z$BW#bzuaid`T-12f&7%Y6oeZ;Yk5-l7Z)$D<&vr|*BcVhbrBcVaYMZZHY79XdJ-Oh zUQ+3aFQ9IHVZuQkCpuG`aT)fwYKrQFuNm)*HjAYQYHAEDhtCpRRJD~0+SwsELx~Y=b(@8)rc3pS;_X8(u$SHE z^Z&0hR=~vEREDWG0+D-SP#nK >Nhg$S!KW#}SLiZj3^{lKhPusb0}Ba_FiI}FU@ z8C$C`YFG0ZonQx05ru?;z1uxRE5KNA17)x;L&Z=IO0`=p+pv<~E!wHt;Dh ze;*^9PH(u&_VFQ9t{GwI*=Y|O>F|xNNxR*n1ZRRo5?_d|q_f-I56-;+#~rDmAF9E~N2pTw!}t$Xw7+ff81t4mgg$#($?l~hC3ugrYq-(IKa&DKrq zkG-71pZruGpVD|an5OqG5PR><#2}2M*>wYNjSW|Z1!4CGaJjdC?|eTd_eaj3jDDWi zzpKO-vmvMK?Ia*(Ssh8VL|p_^YeUre*^pAI^TKg&vPUed@PBa32|+Fr|I4FnHW=rO zX0Uc|Tp3Ojzg?n}^dJX{yAb@kPL}H#uvn8BsAMa&RLozz{(sf7Pk!i`e45)HBmc;P zWZU7A9Q8Sqafrh9;R%3r#}T{Y_+6_{=V+Ts2vfKftEy5ySRVUJ72dIMusuy8l->|* zQakn{8L>JgCVguvAG&K}YNvs-;;E}Wnxn1x@&s%bHy9$w-y6^#4)GwzHdOcFuuX5B zF6$-5I$RlOr&1@cY$CL8|F=@I2qz4bOYTY%oPN{_20|1B3Bn}*;E z?j(e7hWW%jht3YkIoo^-vpGkAzJeb!#U;!8>KzWivPEe_D1Z*Rr%>A9b9eJjp^n*> z=f~e(>!dUoDt~rNq&wyjDV12bdg{;@$O>}IdgM$4rLgtL&cNe9(q~Ji&}ub+9W{0q zAW~z<s9X9y~8H#BSK4Z3?1Os4p1E;FPjeMM_BzHmpe()-L&RjocBCu6Dj z_u$PPyWsL|g??($GQH->2Z81J62qp#eAPzVW^Dj^-!FCn{a)+^R z%vUlzTnNd7Ul8or$r!s(lW_`z4c0C2)T~~tfDNapVi5#kZ`UeB(i$FE*T|$67nkNj z`{&pkiLEfvEA5yGtz`pl=H8#B1pxiBLb{WE7k@?&KSIXITTL6!4hhz}Q7sTZ9lUcm z#D!b*z$9)nU?xD>iZ54nyBHehJ04XE6&gP&MpSbApg&dChOOKFt=(v68Mp(YV*4eT zm2RuQt*T8pxKxC?L&VlaqdwCu)-_fd;j-62{Zp4}K_K+yft7uOPZjUldj%hx7vuXO z34X&_V)KqhCX1s~{ub%)qAcD(asE2-F#qdN-r&@y@>Wt@$rD4GqbC*wW7Yb;~qW7)E81RQMTW&1x>6Z3U=A_SLk-5?oF!TU$1Kp{JdR`=ys zzMlU4AMLh*YtOyG`HU*5uddAMVyJMh8FE*@a}L2bhLy&*ONZ^|)3)2^ph62T`M4l+ z?kYx8Sz;Sk%4cXT?td5F)u~=JMex?<8iKAjF)#Taa5&&_V1Oq2b6Ecdi|Hxmo4aqx z|ATk!p90Y;QLl$JQ#qrcClrXAvcBpi0&RM##S07+|Uv4ifg z&e37ps%goA-#Zb5B04jr-S`D zo^KvY67`VXuDYsQ)}>KSvtEV3c%t{THKU_qsUCg$$BSQUqzSV5eSE0h#^V5ih(}MEQc3QOM z+G&TMGl08{&A*T7gv96J{2n1ph2J~C5Q|H$J5 z`$@6}op3)JoB#9gPy=P$p6E?G(kfI03V>M{=03VKkB|_cP~V~AFL^ZjCEp=XhkmPO zeoJO^r&HK;%_3ivFYE( z1IM(FPNu88UlTrPH{`b$h!Z-+V?twULA+~h3+J?*-{letPoPx-G^Zd!x$%3LBed+M z##9Led43m%V$u+qa9XjLa6`bp&7%{W4799R<}R^YAgfnIV=JoN#`k5fbv{U zwgf>h<>HE_=lQ_Y3Fj@AvSiiy>k}2M^J`I>x6MbVqeycDr^13pla9!{%c zl6shtW*C>tmS`3)04OIg9$R4K!8=w2>}DNFq#TVdc3qu1f?zaNa^5%qk(6#3y`?qIqR@neInth?AE zno*;=sGL{_M(E9&9f+Z1PAbWm&00iKf&`N_T zJlHyQZLJRt_bd5lJnZg2n(J}>;_0xe1d~Z+_uU>HS1Lq7^3Vhp&Cw=Dt`EW9_e?ky zqzZe8+MXVpz@_Q-mZG%H=A5RVHtVg%IEm z!#}cfT;>yEp`T@L4*#khY<4rAf{CtutV_JsNM@-t)>iAE93U>{_H9NwgaG*U@QBla zmaSA&ni_^?N=JtL7l%szEfEt)%U2Si0R)7F61x`{Uw?8% z?vD%H1RE~+vD0e+F`$CxIgVFwo>Lpa11;8Pd}F8|<%X#As%*3bE5+QdB?k})T;V7Y zBXQ-BLv`#%Ip&ZkMkg`Cln^AUA5zMnU8oc2Q-$}cex3A?!>1h=;RUUI#qaq3=c)lX z9$nYSAM>c3@wij()s*f^>N8+CSXplMqp2aSA}svly;Pqs!m)&S9tDq%;aAPD7i%{@ z)Al!*wk{n5B%4GusR3j#`F?tWHh;2rNT#fg$LCE4MY!b`*v{{2(PYyVI_x?ds>pgA zRq|kTb>$K=UH59S4zOO@Vtjqcs>OQhFvov-dNzAl@CTeBY$k-d0(6?i5{t_ljtL{T z_OL<=hS)lGh;RB2&Kx(ksc8aSnmkg%gU)`sT$Tt&naP9qo(fw{{vI0rJGHQ@~6q-`u*BjE^HEschW>+_YqKM3M-XdD|e@d0>B>n~}Wm zECM`LZ@Ku`{q;9_y2O>bkoetED>ZRZf{n(F z=ew*NxFZ(u7GWZV$B)4W=kSm*Pgw&{a{1OLIE_y~dm+UPc3xdO?{U;{XwHbqgDU!W zYwnl5I$dU$kG&aEAoKSg!CCVbwAs=fH=;wR|2hB|a05cGnQzGGcA^2+uglx^ zPv;nq1REO`OgNZd`Y@5vU~yrqjci>4A>a7Xeu-|)Vt5Y)^{^6@we=h`F58d5SaseuF}E+^h72N+ z$N{>bZ+>s0^cQnuIU9D_(MXUtPT)3l{Nablz4|~F?KT2WJ1^U7Mw~CQ={x*F5e5p8 z9Eu;?JrUG z^k%&IL>6h^dyIQQUL*a+fc`i%nY-)Wjsw~0^Zdv|*Y#HWShC_qs@9N!{X04aYbo$B z#6uVU>(&I%sB&0GvXpyhvg?G^6QBC^2hMIni4cO9*@Gk1vrv@r(r?}EFoY^Si;_bhw3>*n=%t2(r>af{yXuj}xHm7gKpL>3lA>=Vue{F^%bT-x$b|a|vTeVTo` z3!%(`wZ)cxLCRyoo=!wPCTX+2`eru|c^B-DrM1EqD+6~oC< zwj^7mB~gVx)uK)xw=gnR_7$plyTpELT8gB%@%f#X3DexD47hM(23=gz>;JX%Ae2pSA5l(K{v$}w)dzLHt>;OlUbPCmDU z#Z%vORvPH>oZ4}b$Z<9LVX{BEMaBUAT^ZcUrcfUUBCEN^V`$TB#n!MD2MBD6wj}ZGOUa%d~!5ogVyrL>me^8wJBa>gl-1_|#X;CP9|JE~_j-f`<>xq*C)*&`7GY|YEo;>5ULWK1-p!YD zFcrnWR7pko{c(y~3;ts-b2k1F)qJ&K;!Z`A69La(9Iqk1zA%yJR8860#Zb7g)Ziq4 zFcnUu_>ZK8!mn^>80h~Lf=?VWQFW*? z{&;~@bSgw>Rjw^U=}QR3+}#V=?FkK}B@eZGneWuRbJP7hKONk>F=n1c#boxFE=3jp zdCLj}-O00_Q8;aX1&NJ$i^6P2sF&P{P24u1`wj=D~r`8_l^|`It6(2P{WLJ2&aJm>R*g|LQD-e8;;AllKAUv}5P8fE6>y9Y~ zX#GFX-N*^=i4+$zD~TY{3{^7CQ){jyIuW;R+{Ifm%mDb1@84gBb_-+zAa=f-65?cU z5VY6*T_!}F8B6YonV$s?+YKyF-p>wa3d)8}T_P~T?NIo3ewoM`b*vJmqK~n#w5G+HaH1XKJuAM7p43g$$pLiYX-H*O4~)FQa0m> zeu}97(*lTipXmDJ7AS`eeP=P30hwt$2`=20h3$h+pS~$P{6J4~&wq#TkHZ3>K?@4{ zv%v1>bJnm%5_J=>s;H$K-Xn=;PT=q8hl0>ug1voPMq%@sH)d8;%2OR^E{$shmN^h4YWqinb9L$m&-%+B zTk@o;HuvS8_ShltPXitej81u+10bZlw+7Y__JIzf4#zdnexy46quQofG((~U_Di-N zDYg#37_(VV$f$@FTjv&py=;r}rQ@K;nL1xl_KM)~@($?bkXL;`NBDAo2^WT&4V#rn z8*Zs5Yl$}*Dh2jn~1*iel@jOOs~uy7uPxU~Kcl5?1^ha);4PNZq)OJekFlpti5sFWAl^ zt3b+}Gs9!&?c%qW5IB;~{wlJi?|qNobK@Mu$-=qtT3#ypmmx0sW(u)j_ywogNgZoIYbgnYieuOD}1*lV#TqOh-M zZDfXE;tu{CLoL49@A7vlmBy3CF&)b(0hZm16Mw0-W)0EVfuOv=Va4)oWHs<4931nz zBPn(MjZy+wxA;#XZjRUFvF_A>amh88*+OFZtBp1E*B{3w*I#_z5Sokzm-H3yR6H1% zW2R$d&|KG+L%%S+MZWhp@dV_%>ySlienQz3gly;PVS1(B$%X4>?dRib8DRP#3G2f5 z{faJa#jTGop2Z3uR6sU_JN(Q`cS`EJr@*tDB0T-xdxKkso^P^x6;;38UKC(mcE+1q zbk2iWs0t{ypK*sfs0TY+8!*WfUU$g(%9-=y}siSaAWf^ILdnOg$SJ zBefampUz5gdddQe?53_T*bQ>^jGQo+DC9r<9) z*5n>WK@Ca5{ZdziV8?S3BWbehWq}ACSN1}|knM^MDMW~d2aKnvX?}s!w?+y} z1*s$)oz=+>?*j!_pEyp*#o-kC4FU4+<3^VAnf%@3Lv}Bd9XJ^=X^!6V@>YWgaUVPM zY1IL6FBE>53uk9x*CHc%>FN{Nfs#0n+6CvW+WbBgi6xd#Yh7IIzF${bv<7<|RiECH zpDhG&%F5>xIA$boxiW*B@lfu)-*|d*f%IK|VSU&r{A&L?_2VzjWQ7|*LeCG(;?uUx z9cgh)q?09x(0lhcP%{84ypBKSzODbp38s?C&K%Nf$^V0u1pU(AqB<8!%t?l4`RsxMR$yovlSM!4OkOB-9{kcV8uq^I&DX3Eut zwDxF1lAf7sbZ~Y5){I`kcXWP3@+)mxnv;Ej)a_GIgsBis0|}Al%r_UmcdWOAZ{pk6 zZIgx3kat_3Bq_tyd*045=CF)y6zDY`o^UYUSm6KUZ)~o-4?FS@|=I`c8|*=EuQUwL?yX)qY*EWfqhG$CS;I7l*59(+!b5 zqN&sL(ivQ+dUv&9=6%aH{sHgY^?YNz!nHafN}sGe6amBkh&$|MrFWczMxAB71~VH+ z7&}VTpOzQjgz+L3hTzxB7K|B91BXWR3QONg|?=H z+Ua;E&bDXu;(|}HZ#!Q*iu@hstCRz%_!W@@4Qw$sd;XXL>*lP;>z_`LddQl}WUlx- zZ`-)w@3m8OXaRTOFksX0n;mD0?C~SSJlwtZL$rImLW^0)E_xAKuHB6$K=-EyH`0@p z)mN7zCpiJym4=fc#@|PJXQc<&SI=afo$g15VCO8Ne*;+nYB)2q(R|%-Of{Z27{$Mc z1!T;>E-G3`Ln-2bXg@yBFYfJZVgKLy=TyYfj7WqFElmUHT^7G}7XlJF?gyoSu-zD| z)ONCLWBS|~E^c_+96T33XTE-I(;4@m#nWx9RctjEGR|Og%GngHyrPfvP(5VU`}Hsq zIJueA6vnXqQ7`Jd?W!-=p-!3qbt=_$HDeL+JP8E@-$dzfQna}*YXJ*U8Gk#VA!iW5 zcR|BFJz(KqK&)Wzs0IKCI+E6v( z3I037H#`!q3kN4vg`MJ;Ftf2ytNV4+U>TCf|3}tYMYX+k(Y|;n7NA9oTcNnSl;9MK zQ@lulV#R~ILvgoa#i6*nYjAh>;4U{k=iG1HG45mD^05DVtu@!2za3-t%xwfko~1j` z>?#QJf75DI^jS_}6o2@p@SfRc>z#GwqoQS!R|>}Ew=W$QrL2@;{+Mw7-@mkKP{E;V z0vnyW=5Yk=uRS~O&iYR>43b8t<#<`_mdE=c$~%TlDw#hkBSZx|lizuZFIr{HaQhd;uk z=sojnFCIMZGezfb@_61FF)6M@#0Hdgsr1t#@DBS+0xj-(I;utLG+NO~Z^7X;RrL4K zjVb%ixfK3@&z($dTstGyBcz|?h#X^Z{x*o@F~Is-3w$d^GZM2&kr%=3%wfNZzf0_H z;=ih*jqvvdrN0^@IqX+A8FIMG)qKC#cHn(%h#HigW2PePw&K7|-Tdap*zcpO+E1@W zBl*U`9fbYCrBLC=R6a;1r&BifJELZt-V-9{SM;8>oQy-^9#X z4>FT$v=-KYiJczy?LN>^WP zEC9L<&u!Q(7qJyWHqDs?<)=&mpKEqbYkdfAJuH@6FB1i+TZETWSdghc(5{bpqX*dj zAdW=@e#G?e*UjgQv}J%jG_{w5Yh_rn_nE35@tdFs&Y6x14M3(k{DM{ECt?Q}%f>H% z9zY`B@Pw+!x5W92QkZXlz-HJkduEkT;1&GP9<5}0U7_UTPwv>&=_S+^eBchtoG+hT zK6JX27du8)x618!{bnVHip;~9vD<$1yv~N$QB3wsF$(W^jQfIdkyw+;QI2)MAN9oX z_Pw4ie3x>W6SB{!2&>YS^Ft#%4g6OQ5K_+^*P6N2WR}NJ$H;y^A|xxU#0t^14fgsF z*DkosKWgIswC3~{69T3DR}W%`(Iv%UhI4u-qMjfS8#a!E{k*?3{#bTTp3J9Ly1y*T z{r9>ABEI`bzJi;%W3#>dF(zt-A<$YAWWrk6Q9p=iWAf4&!~OGoH<`kJYHj` zy*X6w7Z#w+m&EkB-$xI`H+g5D_ryhZVWRBr;yU4H0!yM?Q5Yv ziDRTew>MjtLg+V|`yHh0^M)hYU||9PwMXh+FO$(o$0?&zOrP&?b5dapgb4eoJcXgp zWx2j>zIfHxsD-(v#%GPyKH1hOU=jH`tB4b zE=3K5^5mipF{CVd1_@cMWbF3&*Eacq|gNc9^$bs0lh_fxHHN%1m7Zz*z2+0&mQ74>7*4| zgI$$fo>by7eq$Ql7MNUromR6l-MV45O!3|=F&!+BA!kthQ*d}Ldb3h1{NT4&n%t|t z&IY7tO^@<<)SK?P)I)-F=yP3{I|=w!HPn74*m@YIJ7DWus!Uk9XT}x`575GFxxgej zUR@ot^aX_!q@md42BJMmr?qsF@@ky!4njM%le-Cnr|1zSZCsZ7^eZmjc}A=nbxJ%7xMpO|L$#r>g;Gc>xVH%F_fAOp?R+&?S%9&;ozM`} z8&LD%U*#y5^3T(_)PAQgQs7D6d+mhKGH#|40$J&guqthc;e%k)D$^%oN>%VN^zn%& zuS8QRnu-B-JO16}!S2Y}ZaM{xOU?$Q@r!MEAD;8FCr^a&4=Qs8oyG{x$}}wF zQTL$uyS-|}jeU9&mgko_ zh(Zv@k!+tMtMPn@z~-1dZ3!2VMKjL&1yGU&2Qh(KevT`dP{=f z_R5bO8>!KBanYJop-*;{U@d#uwX0W%(noe`A#^rKMHY|K9!;TL=I%z#joc$)lj*BZAB2J+YZuWyOaer_G*I@!gUJ`{e3- zuu?H;O7AAvUC&|!Eudp+p~!TEkDKwi*B(#5nUmkVT@sAtf5$3Ey%E4SDou;O*F$or zu0OhJKTJnPXx+?jFPmYJoMuBHhbUr#-^-^A@a&KxchYVTj1V`islf6d?-E2_xib>q zj&AZ|)!N`Amcc2o)pTUJm?OB~0@Td5pVU;lc-7w`OxU)o)3)7;ZU?y# z_?pMHcI+pE&CF#cWd~;aU5U!8W?lq24gh`>)NPMbAIE=BTWnPRY+%$~9mT*Q!o2J=^v zIKHCjR6h*60Sl|OrRSLcpIx-w3vq+#%zQ#t&p5RJ8MCR-%Z-Y2J$_2)haMXK`2L%) zjiQ7#-F!k8QFdszE#I(Y(!0=fqZLm7KN@Y#lBy#h>hm2fo52FSwAZ%5rN_(D8g=&x zW{l7uHL+A=Ki;=-_UassK^D3SCiBm8O>v~@D34BSai^J${NWeBO)56I`sV4kCKe3R z%`DuU?`G_lQLJszV7765`XmT@wT(o;b0n9|qBI*758mte>toOt3#kGD;Cx!ti3sl1 zlbY`2=*>q9M6o0ZFx&(h`cE(CDcCyDd8SZ&*K>#1j*8~yI48HEK^Z#}+gpiA4gb)D z+{5>sNy0hb7l=_|y5J@Kc+4{FAL{io@0dmD(uE{IjUYmn;3W^)i)XuNwoZEiCSvys z(~Tb-H7>*3rlGj!46z}t+|ePQp)$1B^=^86 z!$6P-7;F3|K#BC|PoSdLr_0HmdhRPzspKEXP|K_YzjXfbiVi=F5n5^30Ia)CRx557 zCW2y_4NvHC#{9zYDK@<8K7polCmZE9o%8|^Ta0`H`*SJo{=>Wedh?snmR5@&+z^E}PPCczKd5X|m8q+4uP-ZVi2x@v3!r}$%PK%x24~21^Qmv%#6~KQ%6YZ~ z(|%z4A|5R?HE5;&XkF>Bpif;%X%7VqvDzGmm{Fp_fhXa_$fyehjOpJc zYdXY5k`s(*=3f_W9SY3_kYCl1f^)t5M8GQSXu&LVz$o4IasvkA(Ae6 zE^S~|9{G$;bs~@ol!vUTeR`?K>KKv2>tIo^<2bFAV?#8p`RGCHiNdn3E%d&W5;<+d z6bF!xUru<47Qllu-O}J{8@&prlClD)>(U`edNddP6Jr}okq%HE&N}4tx_O}GfdhNy zvx5G0IX94YLG>#nKCcwo|2+L(0~J0?>=i^3oMum#(W~t|yTJz+IsXkcSHWJlHFrJw zVH-LH+N%*PXCnFSQu=+VWk8&C1?jI34_0a@xGr>0izj(`ENFJ!VQ0s{f8K z@8LD-ta3g{fR&UfPOlNm75{BR0q~=gKxnX{+Z@M7&y|ouoFmVm4xt2TMG+g?@BA6d zC;S2vL)t2r5T3`2besNuelpEM~AD{Xk+>S&8JCfY1oT8fY$9Iao>8LjXSFxLYO zUa=3^KPqs9MBsnx3SbCNcwR^e2>x~^L>ZSv=d;TUKAn&gqYWW<_>?Pc-)onRDhGyi zbgnOmU$=zQMt-=4RVKb7B*~@p!PI0C=Lp-ZS)Gv!{9tuR{A6<_H5S7D$mpxR+LG+k0PuOjZsTi?n_i2I}_d|andq6@q>naj^` zu+^R|z(=?;0?VYkG0E#_VtbMYp-s|I&N>VIGZzDt(pA%h!C)HD)la~ zf?Ea^-WJd1o&w(^r(DzwFtZYa&~vkO=M2HyduMIIN`~o*i%*x6NPw9 zwa2=ukAW5|Zpg%HGImFpRKG5@S0j9X8~NPSqaR$;#9G%I^PGQ_!BM~#_ci6p`NuDD z6#i#azwB6ldX*K_{A`swhfv4Z*M3iQ0u4mkvjGDx^SbnWui?i7+9!k__1K16Ax?)9 z*BS)&jnD0%`PB&iTwfTsthOKk-{!6~WlM+gLy(1sl%txN!_9@J@%8zGsQ?7(p0%G| z{ZCWq{nlp@X8cQ0eKfQS!|!yYB-L{;=r1G(Vgc(%NWrtGZiHWV6J(Uti$D?y7lxt# zIl#ShVA0Es$wE1hnv2j_^N)j%Z@RI8(uNYq=kT(1WU$9l%G@4LWtcC_PMtx{#$DZ? zxj)}hl~>pr^AgMWGNrfC5mk>Z$d>nU!7@kO!A85h=&2ImL^f&SZwEfOmP6)8KXI+{ZIb%D!TD3``+e4resaLU(ubKMh1nbFTfVwfhK-s z8Y-MxbYVADVOnr8seO0U5Ay)YWnmH!TIA- z{4cfP%IalDwc*A^*gs5-r}WPZB}i>j54aBO)jZ@D?UP2E56-ivJB`%dWlO?CDr`rl zT(V|p4}N(HyTJ}xU0Y3{ebUx(Z!z`+T2@OZjXn#Ft7B2k74aiq;1KT58lLQsw}`-> zmbaOp<8O^^R5sFlCl=(D93Vdt&RsRkGtDU#jL}^`0CHsK3Eon7E6{%w!eB;3ZafCiEDRgv z@LF$e9YgPLH@JZXe^~h~FSRdH+nU$*x4h~-A~Dgw7awd!mLOjNW&-Wjl(^Tr)LLbC zVBXx3F#8H}Kut_@?5@C<@#wwhrHmmYt?akv%=G~-y zgM60{x1}R@ZBY#=pu-(93=!Qt$;e^D>ZSw!Qe}s24NBgN`=2_DGeiHZ|7pt%6kmE+ z!DHH+;#sW^$U5x|h-;L@s57CS3~e8in28FWQBLXAwlQYN*$!6+VaUx-AB8{bmwnr} zrfsCf9J!cH*6W+j?qp)taFbLY%$fCnUMh?1fW6+*EJ;~{g>JZx6V)Jgz%BQug0+rr z|2D{b848r9-aLaIkiW2X^dbP?!H8kI)0wxds5kuniI!dUbz9B?F6H)-@mrGk@F$+o`dD2d@Umaq%d_kiB=~vG zr1{<=L!foozW+1e<+3ou*M0kY(|0}E>$a~;OS^+;+h>nz+2PGQ0j6bgz!i6K4Gr0! z|Eevd@WnW%##$6BqmVa=p|Y^klnwM@ddo1jtyUZ-#A3gLP9uwt5cqa``QNme&nY>O zC0G%h?B#Pj4!FyT`Ai_7c8o0Ue1*&+duq+7Xcvw@fc=_suJV~xX+P4#21zn{p9On* zog^m<$A3X~*qGJX?hn3_!ZQaN-dx2PCQMI7Nz6f#s^RMrqi z|Xs zR>(r^v|4DwxNH16(43&)%FgP~&C2uTY{22;98$KW<(e%W(%dbAA_X85%@0)b*etHo zva(x7)OMko$;5zG%ciyMLjC5r5us@66*qvSA_2gSH-jgEn+bg7Lz#uahRO>*Oom`` zSTeNlz*l)|IyCB=MN}1IvCsSQfRnmPlOL`*+aTyxqN?Pve0?{Ul(c!5X~iQssc}@U zuGHg-so_`TmLEFnTw}s!gO%M_iTt|ZSu$5bmz!46JTDvet(LaLxkj1Am-0@o*>9aXVD14_wwy z^wuiSSC2Kw$cr5Hw_3s)51U}|rWNwFk!GGTJ~dd)y*dBQ)-r4k+Zcm|MhQ) zHj5f)bOky%%`k4a;ORERz+zM7|H~lS0!Cw|)l) z%3d}Uw(7LVfSanmU~_ApZ5Voy)%Eo3X}K zoszj#z8wy+ohrhE>?a`@-!;tLoN{;nCx`G4G|xxXbE?^09$voe*OB+s(+hS_KlA`XLwD4%6>G zykqHg`;h-o$u=<*;4*#7S(>-gF}!%~&XoV1gQ&0D&vbrBn*|D8(SQW5%$6{t`f;L9 zx}pZ40ONPDFDX9zB|Qtf3u$=lb}rNDL&XNA$l#Z~Xb&EWJ=(ln1B6QOsvypZFn?}r z(44gUU0}`Bl7{$=0t&Q|Y;+PZ=?cZ2(K}o6UFAxj>;JmDe-M(Ll27Wf*c$GFd$*i- z2YQ&X2he2xEb_M7yEzrTdcDZtayrAOmbX!oaJQCdPqPRT86Ye1j>`@_8{uPb=a%2lpgcH!DYGJm2an=Vh4GyQ(AO!dcw zf$E((6*O+8M~|Vt&^MV2N4;k$W%l*-MqHhcYaN~v3+8cP0JT06Vx_T-cKLbFP{`H6 znt31-?RSwOO|J?|P--KgV@xO};=ik;hgQ#L^l@UWgJ zMnH*FBKUmr6HX{x4U!r&V7^*QT-=HrnV@3wBmDTrp)xuH6i6oy|3UnW2FtA<7ve3WfGK%KsEy->o89JS9D-_>C% zjzr=(>m|;p%Qu=({N+2sRLI+$TDMU={w}MU?y(Ch{74$|FxX+7pnQWPG=NxhAlw`~ z?;>>;FSy+Zi5>!XjlR(cZJ}lVow{m*)#99)T*QV+Dm&xwv#EE#YMyuw7cCY^b)vO# zXI?HI-*UMZ9%*mkWI5s6Fsu>qYt!t^sCXvl?||6p=S_CWTN-1wi9jn`YiVGvz8_S- zUW?~LC!1>@1>a}S-g=f1^}!6TuC?J%M;EuA$I>Hguj~i5HT1R*d}oj6wTdfoU9>co z>Jw8k%6rXlSWJ_!2;_*ZBi3D?K; zvtwBmE}M25|I|i*RG%bT;bY-}Cf|pG=;Q16N+NpIVMuJmcd9ql5%?i!L9FO+omLI; z%gQU85tQ--TlL;9oC8vMLom`>2QzId)&<(iHG1MRdsYgFeul_3jb2|}LfqmG?+r+> zpL3ecv0n#32io~u1i9Tk;UY@N4 z>)$bU|9}^-YJeWoNnJInFtduRM@!scMp2m+st7SD`r7=jGOk;#DBBCKwUygQwv8dv zt}0yO-sS26ar|#k4cb6o^sK~3!VsscF^i9GdR7EMx(h~@9Ez@(g`F43*kbMGhzlde zkVvok^0(`D3RjN$z0A)0dZZsj)}>0b+XLX_(KkkLJMh5I-Y7t4-za^3!7l3b;S*Vx zOcbd4>94~u%e|riqN{BpxDGEHDytwCYqdzw|g!)=iT^jBJGil$)!^`MvB#Zg-7yo#37f%yfHar1S&tf!Oz0+ zz;3SMqE|#bC;B)S3QJ)j!TD!)32Iyo!y(JKP!WGrM`0+kJtCt$(lFR)owPgTR)V^K z$VPVilel>jqBAESlB+X)cTb~fowgP}V`oxfVC_f3&J{=)zK%ub7nivUPn>xvYo9k< z+(;lI_MJ10&|h1fJ~l7k_czCGcP5KNI#A16Bx>=$I{+9qhz`*(;TEyWOik|{%H2nr zx+LjTRv!ksdxxSPoG0BQkIeU2#DU`U8TYvdLpN4n6b)201>^l%l3gGQnRXZN+!anB zB;shhe|7n|%vCL`o?(|p>$m2mvo7u^Ri{qw$wy;MxXKJinyR2G(}`Mg1JJ7`J+g=6Knz3GmI8R(LMb8Jn<%yEIx8PuBMg~zYE!PG8TyWK8nqW0F=hda1Rckz*KHgi+{1cxedq!#M7w$PPwQ8x`H( z+@FFTlBi$Vgch;-KNI9(dt!aP)q#hopuOa~Y&!0fJ}=2`5za&bDAfIEPF8F^P@|6= zUR2j{aem9o?BD3;ZagxInhEGryAeH)iu9A#72%ldKx>1^fvZ68Sui-V4%aff``e(X z1lRA9&8rw?%<@F=LQ-(vxbyc4lw~^cI`ek2BUNy^={bjB-~S3%2_!kz`W?UVp4RV~ zQhF?03eevTI*h}!5yoQn@^Wi3Bk6x@j1XS%&p>yI>ytzy+_32AekqPQ9O?gz= z(Og`M3oTQVxq^RJ84+=lH3wH}_!@yQCjE`C)#h`!xn;~quQqvOVnQNKKLRRXH zgMCkl&TEWrjt3@8Xo^HuWPc8*Bvk6L#(&24t_+*?WE9K7$`uY`XNYL#9Y|=W`-+^$ z4KzQZr{#_+Qp<0ye1C(i{eD?`hiB#2Ql=XL?~)D~{TKXla}d6NXxHc>ng$=^ngpgw zqtYQoKk1bmkz1H<&uy225%_gVEjKgSSI}30ex-oI?lGZoaXR#F=NQ$oO(@l@pVH_< z=?7dN&|7qKp=6=Ya|FX=O!Ak9haF~W2k12Li;BSIK4XLyZ$zgqP~gNa%7b8+0LXOi zB3_S`#YY1Nc39{Ue_w!~k9cot)^!orXMZ%?=`ISsVjt)7wA@Jkkq)sry;!HsT-`A3 z*&hPg^H@;N^@RehL`JET)MmIDcGodpk(9X7aRK}mVbQEXgzFmLBgD>6v?x1|_v$Cx z4Sgk%4$8m{8q8h8kDnyNY{XR`9a%29o&s z@yU-^4>FhiLO*TKM%3K<;MHgMm*xW!fM2wj$wfoMC$grR7RMU*2*dxs-b#tJ^CGqC zypKqW#BVlYnK)zqMk#!~^#R?~XZwjvM}G)P!U`=Rqj@NPs zeNzpLnZ9p2#TWlwa8&wwVm9oDr5O)gtyUO_2|fr?0T&)&@a=52E24o^f0F$D?{7va zUyx_E8{|G!sZaQNyCDMk&--0y8`x1%K(n#8P2E{GE>&IcXx|5D1FZpHgVOMtEriYk z9<=h?y1NS|PKIt{wY?)wniTImwDV1)�NVhQwCYtQsN| z%*kd!_2KD2HM=~K=mBv8w;v{}i&B%HvN%)9IgfaUJ(E*!KkK!+S@BMl5|fayzEUGV zl*}&DenPsYjwZv0Ro;fuj5 z#PPRYyv{D4eVxM>ohLPUUv|ijJz+R%L98Wu=eqD`*D0Uc-{q4BAN#)YN}3^s3*27x z?R|em-VPcsN#V(55Dzf`aD}x=_O8}dvANYZa6b$B8VizL3MlQ8l^i#43ao6VZzt=} z^GQ;{!@3a@@a%$p)Iz7&$%=!2gysxI2yGp?Yin9fW_Cy>V2#irSjH^~>@E~c^+K{^ z{|>u#;;4orM#*krNGUAjQBZgP`KR*@`C#I>2cef;;y+#e?uVD}rcAqvw;B*87iKuD zsFBvMSUO8L8#pfkPSN<2>2EkV<~u{ z=w7+<8)op7U4ZW$yu`E;pWqaGUWS4a@tYMpyHwzk{T_qD(g5%$fS;nfLEIokjxS@V zN;O<@v$6}bfRbwrK`cx56{2%Z{STdFkznRa(v2&R(<_~~zJDMp&0@b3%DG?=x4#aC zkACeHVt#uDrjyyi<+uqg4k0)z_KIfw6s3QM`Z;XPR5q$*{*ge7fVjB|znHfU-dAM> z+N%F_)7?^thCgp}wRqc6!hRS4(C|JOI%n3mjX;4Nn(r{-eAc>YAWS#z;RUm}8ACFt z+2zlau2v~;L*ko<8r+$Sg=08dJSh%Fyj;a@QT?-ks95lbm(#nJ&%I7UR17uTwgt78 z$9&&cpM2Kwgkd*bCbSowuZDhXJ%|x?=Ny_sCL_WUp};gzyT)WzR9%@E_A*1h3p5{G zGB!SL3S6n%hTk||5#fNGI3+P`3 zH9ayhu;JfZBUrrsC+`VEv;c_>y31H&Zrv4w>eaac{wirIsFZX4`;tdaUuRhdqfA!9HGgxV~C2Exv~S^ln8&fwwv z8^!wt;@_l*WKK*KJDBy?yL)+yH1g+S`mf4OTOH&ln z@hzE}V^v>|J;0(@ESAa`6GF9+KmZb2dc6!j8&iiWuuV)DfaZ>e|WkX zbQ|efXXR+=#O3^diV)!j?#J^tP*YvbSF#-1*j(?o0eWA{x7b;Nsk%u%eoopZ%#c!a z8z{l}9FeEJnTocR9 z(konU4oRk$7Y2#YX;n{x-!}U&9U4sWrORH|?y!yM*tiqpkKB+Qn}3LW46@*-%QN^7KHO2hTdP5M);5Z+EpwHkb^iuyBg{ zh?GtZ#3O$1EE}eU{RB9Qgs&FEpg56vdaY#CYh{AH#`!<7aD@g_siW{<5*!>C{3%1) zSeIKncE)UnU$zQ+!0dPinN8`jj#oD4K&h)h#P_b`Rz}YU8{uvvQ(aG~O=osz-!e_O@EXu zJonA>yGa^dwGdEPt{HFTXPLO5l*8e64FdZ2oy0rPM7eI@7_ z6_RO`@qYvHkV)UHDeVI`TX)FETjZzwP+le832vH{r>{8# zs$){Df?$DB&{|?`ir7f1n;+YH*BuC~SBz*xBna>1Ld}dXDIDc7!d}d!^c3^0)Mo!~ zu^y^VR4z8x^CT2vXGaibfZAU|UXwzj+$Ui#LQ^WaM(wuNzqFj#rd1@qLG8$hd-yQ-vb;_%yg# z#X`z}Kq6vuCdgzfbt4V^hum~y@;IVfn<#YcxpDU$Y{3yE+oN{JaefonmkA2jEg32> z0kJ4ntkAS-=lj;HAMFd+#q2T=x}1=qe`yXg+oPm)x2X=~uC`#AA1S`w@zJ@kyQ&)#UP|~r7_+S}3++uzmJIIT4HNPB-@*HD*YZw` zoHkeyup5Rdg=XpP9(^&%WK--hl(WH^05)8j08G@ z$^4Y~C6q^02+^&9{!;X*u`m*4_Q#{`{ha){u82f`%qRpg|FlN+V!5lfgD*aR$^g(J z{HWJhUO?VxJjfKIrI~j=X^VI6V#9B}&7-?UagYhJ)?qWZG{>{HJqQ;Aa3nC23wYK` z*=F!6BX{;!7M=Mt)v-Y7Sw}7v?2qDOgwGl2jOI*$`imfP-Rk`U(&`0lJzkWM*;PZc z+h+XkIr?R7pM^=}gFDR;k7>>!h`aS5fXt}cXW*r$YQM%O-ALh!^WP=a)RnifK*;V& z$LnNF7QINXJnKbbrvA>;erz&CXa>!FG4yhFmNQcFV5z-kQv+DzP=!ab*A&qZl$2x8 zHHjtfbS}ZJ%P$KlAybIpc9+;MPx}?~aIG}4)ZkJ0aqu<#8Jl*;@$&l&?4M%rU^ko?IS{s@&sUu;8ux$&hNNp?R3F%+jSZjQ z8h`^r*ibX`WFA}H@7CEpK2YGaVfA5ZX=PFvyeW8cZk2l-$1?p+Tg z5Zju8WDDSxp?w~WvZHo@fE|Logd5qfv zEdeJsn(^#hB`;(d&eGEELO*P=V|2Etr@~gHx%F}-_v&8$tXICI=UDY*qH#>Q<$E~~ z5m?juqq-{+h132;nJW30ir?~A6=XnRLQ;h5iq1rvLs$rCl~rt#u|r3&t#?zih*>?m z%XKDx*!mi@k^i=(6oQj2fCI{tBcFwpkeP zI@Lr$AnA15Sdy=FD4TBo-f2$+ArgdEH9hl2*k2NY-2Hd2sJ&{8$|dduL2wMkoF%g$XXQC_l4b<67C z?;C%y{3bmDk%{Ll7v)?>Qu}G&DAh);^{1;z;P==;VBUhG*2mvfz>yg^CQ8z_4juAbXHWUrTm8?*5C6r z7%?4+<#XGiE0}Oi5+$p8bgsF7mer#3Guz_DWjsudG-jq!c#>Q4BV1(~q+W5u~}{rT2$@%gKb%3+;EV zc`d^)ZnUP`H3t%#ur~qVNqD<@AyptZFlkj#$YRr3A1GhfoijG5rJV1UhK*7|zeew-KA075?X^9p=0`|4TAQ8;?@u?vVL}Fo*}x7a?$)LIz5nT*MK6OfN9>Z-PbGWotFpy$g4%+V%Ep zdI5VOr5rZ(ccjBo)w$_vWCJPsb=cF->@jWx3#}c+@cojSadMEpw-1{OQUy5mzk*2b z02F~DQAD;F{R#|S#yS{wcnXM^Ef35j$9?K-JTaS_k@VWTK=e85d*EQnSIhRp$Cjs2 zb1;B-Ri0|6@ZLmsGdQzi9RJFWONDayCoEyIow{>(z+FGQ5elZ zf%B(MX&qfy{Q=Arb%no^VIgnYfk-3Td}baP^6wg#r*^OZWwY{1XSQzx+Xz1RlXJg( z{v6e|uH9tq=1I#;02&$?9?6zdgFid%G05m=(=ItzM@~*TJo-w;2z9tfbJIB)srZMj z>#~x%yd&;cI)r;VfQz>e!3-E`HvmXd@neb1uTf0trf5`V0+zM2DPl4GKUupqD15m} za5iPZE$rm?8U$~4msPxiKHNSyw{@+Gcs%h2^;MNL*RCyC8jP;4$~_z>Uu8@Hm`*TCye<#j4Dg&wCwz|;) zsGSe4XBTwAwFQcJY<$U?E#E4%6|3A7TH8raTfbrEmP5Qs3R@pZ+#nBqO9-W{i7vw- zh(3us{v;(r^Rpb*g4H{rl7E+NKBp8v9?ng)V%hU*5Sp8^k9xx5SdkQu{&t4sJJ|Iu z=hUS`Y4g{@awn^bFhoJ+NFw@w0|pd3`1=NKIPf&(SIiHB6xF96vOxjkpiunB&;A#F zXS>t%pf;Kf?|7_`qIUjx19ND&0f zt#(r%`|wPF7>XTjtG>dJPn9L_(Y9_W`EW;5*(52R0^0!fWYd8#@0DlJm|aLRYfjAW zAWjE@K5PF@P6BpKYGHt-dQIsplT|tP+x2nW73Iw;avkPo!Bix@gtfExbB4PTSjUYY z!4DHC{syF7TT?6vW>Ab3y)fIgbg7UOwu%lxZhf6f|5~BTsJtSFs{1GkaQr0VpHb2KBd2OGlf9C!JCVLtM%8!|%4<*R&CL62T`20eOrf)v?|1aCTCNMw~S{0{wk1zOU$` zr1$({F%E5Y&o7{jv5ZPntyQb>e-UkPq^K&)1E!2fN-+#0V|*iLJYKniiU>}Qu_ z4UzcKAY_b9Z0dE^EcyacqVz|fkfW>})gi-q*X7fiuF0R-@UulIqGm-Y?12mNekQGd z=CLwssK4`qe+5?Tx<`e4BN zK$sbZ6p?-HVbC9=1--C50B`cp*8?+B0oHkGn`@*4^yVJlg3bJV$n^Zz(t#j!f9awA zRE_m`pna<}j(^(X@FsR=Z+++Z@y04YKmkDiRt6sWGZJqGFhicUd^a?VNY^`0!Z)?= z_G7@^zQ5SG%aW5pDbYdS?LL$$Jh}-cY{x7w*CIB(BUZ8B zqbUtC<(AD@Oz9LHuks}KFE7-xaapjhq?lN9)rzytOeFBPXcYpPW&%uBer z!`QZft|89`Ee|~I^AfH0bvRaSBK8$E#?URQ`6wa$7L`*D#I{V9MU1K675tu_n>WRW zb#UCRI?ZUB9^R3E5|_+Lp6XcM*6VzTRD8Mo!6=BPBey00Q4|1xy8*l-<^)K&#Tnjn?vt9)BTG{8iY~wnL}z5#a}BL zoKbDM`oGn`uFdiw2;Ak__}tKAu6};8csi3X+lUkA&}rFD(lkswd*>U3kq!RwlkUBX zosY(#T|M56eG(RPmq~0dZ}SH}A*Qn#vf}fr(uBC^R}rb%y=GEom>IX$U$6<3MFiXO zds|&%Yu#LIXA)V(k>i9Vs!p&OR-a{J?N+s&e+s?*_4-!JX@aY);H!Jsu_QCPErH4Xy}NRz7wR9ePd+|7_KZ-# zA8c##yf+>}s)0gF34Ky``81EqpJW07g*dVK)oAzjQ!MNC&KJ<&ms`)#8^>ok=^OV} zXw~?e$d^BGLP2nDj3R`bTM?x}bPcx|UD_GdMpQW03hjZw$M9~*>qosG377fIDkObI zkr)9&*THd_WnS(W#^PVOEjRu+d@)n5H0FL0>~mXy?i2g`#?>Y3$($ZTe&*V9b9~Ku zXXBxL`vrPQzUto~;GWSLb2U8+l`r+ef1lcwd_FQm49(Bl(e2`)ZZ7UIOB8AtY1nw~9a1JnBHx?k$1$T(!_0P3r2+d#6?O$)xY9Gs#AN zwaFgk5O+dihP?3mtYnA27nd2$z8t)}zOu{Cjy?Fmqc8C|rg4etSL3c(q7U~l z*gk@Q@`t8;XIouG#&@bvX-BXnEf@et0ep*iw1c_T!o<)DKOsXwjtCDTx5UWgAATjQ zpOxR;i`6_4ME92RQ;8tlohxgExQ^;`x1g=bR&p?H5p6T*L*;uzY9Y1!P+^nxS&l7Z$&>Adf~YwpZ^vlf4{;7`t}r)t-(UHechm1|wR z-2K@7x<+xzea^FY2P{N?`DxQG&}(MfzZrw_y7r|LTH}?Z#}}1lk$~C?rlC*RO^}(s zFj*?fUv|l{JdvP&b^Yx~&|T`g0J8;OTh4=)`z=p*lxs_~iT=ZGr4_i;o1S!@Z0Bk( z#5;lX$65*3D%{$`qS^V|bP0Tu?sY%UOI-HRm@IvDFJC2l`Lk?#^2rBHxL~bE=fRw< z%6@yish=L(Ry!D`X;HVQ3ESSjX#QQXdy(@;7Ur&Ja~wumFr$Xhw-^Wc%Ka{Y_ZVqe z%qi>jDYCZCi&q*;kdx)6?&hsEXOj^hm=PXO`*(2XRtwPmlVmWsfibkqa~v{+kl51L z3K94EAbDRk5K&IS4vT|swBs@Z5|Y;)-Dz915_GUcCW{2$-WsM#DN3_fs-f}ojp)DG zN+`5eAV5>?_{%S}$JK%L;g4F~_KboxncZ(v?7I>x_}wHtqqoM#ZX|OXR0Q&DoTBC_ z_gejWKB2_N(jnr?)6E^MLlPe@$1SchZVJ3ZA;fsgA6w4q6e8zKGLDkYA~KX(?3`D; zeYefV1=P*XD2Je0pXv$K)vg_VbgE<79p)Cig;z1SXwKeecE-@aG^i0(-)z@uv7@iG zwZ_F8%1$p5rs!IUvX94!(hRlJ@vG1N;J)~pX=W$LC|r!VC!N!o5+SorK-wB~S+7@6 z?7at@vhwR))KI=oHw^h@&g?m~eae7~YB4)g6*)*?ewm~8_H{nI6MiR-aRA<6O_YDN zOYC2>c`o^TxP9uFqxy(NJ}~>Ljh-U+XlgC37X3t`gZ5lq`|N0zs;sTswd+EtKw@w@ z@YZ;cHU>NCA}0IXBgOkCVp^5ntGg49_QC?bgg(oLk``(7V8Q6&o*rdo8+!US=ekzM zA_13n`uC}mVJ?oWS5zi8nP0tLiC_3eHeUjN*taw~W}Kd5txRXj%EWQ+Tb<2!%TXc`3BVpYcuTg~zM`}gzs z$WSxyrreNUfI)?e{jbHe*sif}N9colmd+8ju=0e9V2?l&bdLl&oY%GfyqSU@!xAAa zi@&g3iZ?c%S$u3T*yNPtaBT~F7A@lMU%&0)Lol-s62Bz-3JMT;2dS#B;{qa zQTf^>w=%m0{@jDt$1a?ApmN>**cw#z-5%cb`U(yh^eO@+A`TH^$W(p8m&2Sqt>4Fd zYbyiNAo<8m2+dfSJL zq^jUx_uuR6<`dV&CqK6G=mWBdkO(JSumH{7n5X*l6-kfjbtqkpZXIOCs=}N7;c*R=^njXxHml=Qd!@Uc8 zUZ}NU7x}&%)Vp4VRRujF@$t1WPf$B@`jL~tvAH9Fs84XrdnBlo5} zJfcQ?x7F%bkj5MkcgfsoBcAoE^hW&mnWzXfJ>hDWfopy}1A4wCetp0l;e#-b)RqEOd!{ELhP?b&OW756)pEdL%M>=vbPCZ{Mu-1fC)7jU;@4<}GwgbQ+SKZi!V<~HMI{TB*`7U^*(0k# z$^2Rbx{%v`dIrVGd;eF^82k+yerK%nL7ZLA+!xf1bVw45PKw@|dg2_^IFdC$1QT*Yml_*P$=&Y_ZX<<@yBPbVsuBJypAXd@EV$EbF3`LJkq zyLXc;p-|&jqTQFSz?M-o! z%~47k1p!n6X0CQ{wr#cXniRB}zphrQ(q_4x&kd@4LlBi~5r)_$SvSNaD44*cr;_W? z=XTwhH$IcfP;&OWnD+XSnns?jP}i9r)M5oJbxXAZh1h>~8x+g!KLPbXGb?6^{(&b%_mOA~4oeyzy0L#p)Luw)dI?J>O zZ68?937rD{o6c)hWYwr3I!=oq zo70Sb$=MWuD5}v`FatBfSuf3F_dIVn9zt0$nk72hv^*Zq?YGo{ueZ;~lhA1&)9;rn zV_thP;By3bx;%m0$Xl(SRgsDwAiL4|0xv^+#J+`=`@}y2dY%C*?%yMBFHWW0Aq z2+9Pn&WqnaYl`^G+wbN!TzSk#IP{1>w>W@p7v5{wypL_@??;Y=&%FHCEcK} zY2U9Kt$=cVQ`2sBv+_?=QOtoiVjE-^-i(g@ql2TNgcZ@YCZyrsS@|#rsx;`slX4E$ z!FcNX(G%1MeKrF}CGFsbNrvmB=Q=A(E@6PMAuXu-rkI}%sa8l@uiJXS_%x1wzYSnP%se&8^Q-`rsj&dJB~8uyIv=D_ioRa6=$j!yAz#=64cr@tb1ZbL zlqRoL*s|GF&-HUdmvX3Ca^ZZ^2n<=5ko%{6<*RSZO%^U>NwimIfZ26qlYTgOi7Ne# ztX4pL@HJk12auo&UFl-Kkfp;sf==Q6KTRC%25V?7f50Cl-JF2g5h1Rms4R4SHOrH$ zZOxBkhmzSvsgI;an3YCt*a2$eQzkV`h2bfO{*6oi1EHJz+`l+~tm38akznJwCn7&R z(w1-{h6N!GRu_QzGle~ri^aF!Y&qE|H~nPlah_j%YSUSgl21_|w-czRt_Hj(@IHRC z^={quBnfgNIZ=Q#7VuEzgtT~o#H<22ZL)eRGVNSk)v$`O2Cye```)o!?{?a-bu(1& z1Cx}U;e?lV&4X?69mG1nLES$Ne`Y4fu7}T8^uA` z0Ua+f_pw(?!X2N}Ve?KWfc$Huw&&vyET;25U29`@L=;@*NC(;0<92~Z=+=PegZz2F zZ}!~ScYYQ3wl0{tb5Afyiv-7|uyr-CLP-VLH|!|MCVj;EO!C;zAwFLsn+ioY(bxikP zuVaW`WIzqOx=MD6_egIIONnmB`XHW3=QKT==pv4^7$AD+jao=2E|R7%W58g}u8qVk zMPFp;*j}%6lhPQGD8TnZ7AfaC+aj>QDPhg}O3OuP zmrv4&XmpbFd6)L1sU9u0*_NexG-2cFJ1L*<^acDGC(I(-+IZqZg(^y~=+u_Iu;Y9< zew?M&#B&2E!9tjCEOpa;p^_Q67y%aXP9x}fLR{?7v#X;LR)J{HpG~{ z%$2@}Af*h9yBlvt=)T(CBO}sVCFJIEwq}z%aKDIt1XP<9QHXo{9h;Ir^L#Nd6*-I9 zpwe;g93kuw>g?Y51dIoha!}@oZ#yXD6-@$6=d7<5b2!jO16C^`I#(3gHs3eJ^SKXB zoPrB(TITq3M~yA-nkQQAH&|n7E#jMyEwG&j9t0?i7kuu0k zXS{AiUjMS5%th^dy6#e5nU$TIQ`X30h$kFC1^WgJKX+Qc0437QI4Nj;KlAZZ@!<=* zfuF-CxpM|%5JY72FP^kjnDp1Jeqg?}3Z9gbUge_9PmXi?<1>z9Pxd33`0qu+GCu6K zrT6iL#jMb84SfA$(zt1!5H`Zo;JXpS*+-64guM|-313eEB&K~J1j@5sAm zR$f|c$x;!z0!bB<79RBHsF6po_^4m3txQ!38rF=kfD5X5M9tznu#N^=N-{FVAX25N zs!-t|kpz10J9EZQD~mYoWLqCKSRognD!ubt)^OK-WZWR9+29UBL1_Di-P1r8LX zw)Wd=eNhICs{c4~;Wlseq-TC+cbUvP5r8a6XKI09@`*Bq($Z6#p@E)oqqPE?J~Zrl zto5^_qVQ;ZJ8$$_nK3KyjL5m;L+RW0OTUjd#s()Irlh34b~{@(EV6%6^wsWQkIdf9 zO@kfg{S{2ezT`ce8ecxj4f^CLpEgQgp<|yJ`^jwhw~zga(T#jt&Y+kvHGQh&hk@`E zRJ>;bKoO_I^^OGYL}`^q2Hym3yX~|$y0hzVv+Fyp zL?SzE(cSY+(0B`w6;))#w!b$0>9VCFDA?Qr|@M#8(#CDMwJ zic8?|)AljD^e?QLs`2_x-?|D%O34@t1TELH1ucU@u^w=-D|_qW&PI_X^}f6KrFQg; zsQVi=7J2KGePpjV_~S_L#|)Hj(>1PR#mm$ibSQ2}HQJ|wwWoYL?Nx^X$8xc{rYM^4 z$H@bp&G~t{zVjurDXkb>Epz7Gz6z9Z3?DU1TX^+G7PW=0CkEbEBSd+-RbGx!)RTg& zZB4i5bKvdM&e88iU2x&Y1k9M@aqHX%5;b?7S7w|I1@z~(k_~GHYO~`M@|J}oVuEbvQ2Mq|qt1W1n4Oc`&Euyw%UzXh=<`*HM10}e%5Jut z{{hl9w>X?>k&#!IXKnOk)zqo7f3LEHf#CQlDeAQSIBuCG|I)zh%n54{9^*h{+9(6) z!R__A%|P^V#4F!t9`9G>^3$tTo3BZmWKpa{Xq=%OLm~E_H9Aj&Pzj{#YRZ*c@Zvug z-^Jj%cdX~A>f$1i+hY2re|y;Ng$vJ#z-l}`-Er}hTy|g^_QnUJa!BXECqOzJY2|8r z;6rzJdm1OL@t^)}{Y&iA)nYL2^A_*1U+T*T+oZSR#-%;M7hCjHPqc3k`nRN>F|!|{ z@uPo!3;Q7ZBy~M!AS~crX_ec54*oymN1z*SVW&=~z|Csz?gw1W#+CeFw<5b1^{vqd zIvaB2*LiO~-H&|<0zn8MIrCY_z+-y8fb6Xn>R6TGWhl%MFk z$y7cLEq3qm`TaOeFnhnS3;<=L7OixsMgAln)Hx2c^sZ>U4np zWf*Fr=^Y-z7<&3rxISp^aVw?`ma&C-@9`%ALptYb-!OtyYEx#ELKshEn8i7w8l532 ztpj|8k*!Ww}md?&%a3uiSdW!}S*UG!T9_oT&!Rx!w7 z*4B}e23yy%(L$y1HqN-nS*S}uKXmX1P`YeYE%H(u^%qJ0R)}yZcy^cUoZHvNs+h^d zT=BAc^)$3kj3Zg3Xna=gz6o?PS?xm3G;#x<6{Y2MF3Mh$uhqJZb$%0agDeTGNCuLn zeSQ6tl6Ot%ZKRI+)9TUu+1Yv_uHnm5Qr&pq7FBs#9O(4oHq>oAFpS#%5#|u$xtTW2 z@A&Bs%+E#WcxOQvciQEKyVo#w?vtod~weu1jhN#A$LY4M!PN zo^Ns^`SI1+J-&smt*5DqHk5 z`dTJ>zwpr_Hs2|wti3oHE_aY#fE%3IJ_CLtN25&@x4*P&`7KL4p0*3~1PDXqS&%g%Br7J-yt>zei z&p2UFsjBxnBvOXbZJ565j&~-h$yoORM)pwDr}R~mMVL$nZaO7n$@(Rlu$u^_h6Zxr zEb0=pNE-*b-xT8(HVU8z3sGY-k~T0<{sb(f^Gc9vwb6Tz@0?Pc9r0?-BXX(n&L27? z00Sby!SFRfHj-(~nPWh>)pD^lHF1%#4^4fT8&^c~9Js>|R8c{{l*eHPeVNNYGZioA zMh2}}!9AO7oz4Di6k{Tc9DlH_TI*b>n|8-t?9F#MjS?*uxKoS?bMQ9y-UM-O@5xln zuwlOTRjEK5dj{xJ%;!|IigRw2)spDm?vmN{(yA=Vq-6&@@U9Y2HL=tBh0@eMwOF_n zxQTOnD4P<;-Rq1He9{>f`3EqCGH#50FgG<|upxjEza)SK#m{j&S;=KGC@ZZ2q&8jo zMCinHkw9Wo*?(tEiGQ)Q)iHc+Qml?nHwV|x(J&PEjNjU1jk~jJa(oY*vDsgnS0zFD z<}UBOdr!*<`cd$wsBuPa9%idO9jQQNWN*ear|gJP`*;1D<}K-9MClU`h5jeq?wGqs zX|zABCoNso2#ufJeSn?nnwttCD6=NH*3J-tOrTQoL}dbUoLCr<91EfGrSfL8s@H{- zbR5@c?jQPAO4{{N5x42{@MNC;gJf@hTw`wM8-EZ#2-VUhWw18e&?JSo2Zy7hYSGQG zfwkvL_l=h;%e+?DwWtyrJbh@s%JBIgn?mBd&zO5_sNE`Yt@H&KpWsXi@jJKHXopbX z#FJhr*V{`)G(OC%&;|STcQ0m}^dt~RfE@E2j3!=@z}Fy!MTcq!_3T+r2}oo$=UTVqTzJhPwL_%Vm+ z%X{7(qK%!)ates2PX5St&NfxIEr-{ZU}8cnU=cA_XCZwD?kO^vVanc0iVU@p!R4iz zsk36j%hPOm0#oF1o;I%+^@{n&vr(QScEcJ;W(uJ)vj^8h<{>fSNXi8s`MRXCJP_^A zVj6@gO{T^TRWS8YCMF2o`u)zGx~|P@zludWqxHsJ>FN5p`_1%*EkMuaQ_%WR+s)%+ zU40?F^2ej8pImiYn^A5zE}e2-s__&o_h;QdT@2HkH}dm)BI%*(=Qo>NyqcI_k43P= zrQjDl`JFj7F8HZ*V;C~(q89=zR_Tr3HJ30aer=-rN$>ip!~NnZ7c2C~)H4tIvTDw- zd4Z={>eq;4jDMM~>*MIHFl0GrlwuoLscVr7>3g%AEAXqTUQ_Yk*60{5Jav-*#H+ma zdnG@R|7YJI4_*G74N!KD_A{sg4rzC*%b+PVedS z`nLSDi~h(Gm~_iYzgC0x&%xl(~P)0 zp2hKt)KBUSONkeJBHx}R{#;JNh@X(GqJr0Cya;sA&Y^q>K=Bn$I~(6oiUS6tjjo$X z(rF;@0vwTMSnbC=6#CXszNzx3S<-6=5)Olz>{Cgn`Wh|v)x z*1E>IQW1r-a6pDwa*OigdyJUldn;q9*>Hjjm(AxOCaJAj-KC=FfQQq~@)Mo2ma#2+ zLU&*HU*FcsE>^$FSX~C{5>3teeO0`ro44TP_RVi7IjrjLh%Xx__BjdN{q1Hw#h@j( z`AlJd_%4J<$N1p;RQn0Wbv!v{hT8ly@j5&ObdJ(a$@mx3V?HN%izDq|kxw(6MB_^V z=<}pOM^muMGNLwyEsXucMCfSQ?USpsQU6Bx3PLwg`o2>A_McRnq`L8Lxk*^ST9v^^ z1Ih&Z`Y;gmKoI$+peMsG$&UYikFHVDA4ny$JJE}DE8mJuQ4fC|$v#!ikj6V_ScxoG zqfplXs}<|TDDy|Nm5kWLzfHBo)hJ=@(O zx^^K(?2bjg=BC%xUpqVOiM%DCrNq(hCgxu3b?ILY_C?)JGmTGx3Di(Ny^@6#S`hzx zG$8R4l*tbfTr%=wzHTpvVSBQlg?+d^Ts5Ka(6RI}23|f(ploZ#Uj25$5;q!6$?R=7-Yt~26Na@F zr!o!V5&33G&+!~XOTTh=F?__?+O+dP^ir1uy)oFXN=lsJ!u|qww)Ti$tC^PY?cUuI z@odvLB&6*#X&aKubZzN_np8X-A3LC23iBR{*szH=4Mk!8OoFWO2?%CMBxH&b00!7o zLeJ(7fTP_5ft@U)EKkn}tm1RM@*o)!#Lu~v)vVEAgRw-Y^w3Z!V>`_I?mlE_pVV7K z0wn*DJEhH}oaD9egF>3v3gv++STG&HloHmaTOzD~4j?U13(9!AT#RKMv%T_{p$P3E z*x(+3JwQ7U4V3&@MXzuwM6-?~HGvouup;h>Dg^e}$Q^;qlPU44-xac; zS2JCHlYIL1wqu2lzvrpjscvYe;0;j@3*iMNe5!$;H)F6rcSQ90s$J1{8r*k&m*zmB zt7Skw53T!%$8?E>$@KT8Dy4Ku3FX}QLUQqCK3R1p;Kr3~myV#PHgLiI7r^`l-4DRZ zKS5y}JN{dqIcCX!FXM1?vuFL~5Y#i+Ew02!2Z#k!b|iF~go&rk@O}O2Sw!$|H7zc% zOQLrZJY0%X%x_|%#UlQ2>66*LJV}E{4A;5$P*UQU_si_6o|OWK{fK4C<({4E`bz6n zr@wa@GN!{ut~XM}T<5joAjG&Z|3{rf^L{fyx$#u2lh^n0mq)%^LGEDF)}c6=$oT^~ zgMJGHeZ)5>KM`|~{gHixKK6xno^#YXCNwb~J{4rZ-zsX0w$le=_{~{xDY6~B5KdGy zu`xe-=5mzS>}TAbbiL^g7JNm7CJ8V2-eNqp8C=_qM|x9_SN#unNTZLo((Xqv-{klv z`E3NQl$EjyCd7Gta2AW)^8 z2_s~pPl8nR^M5f{D}NJoO6m|ZPr&B~lAr4Wk8Zf75`$}458)(WZ4eS~?8TP!kOY}_ z^t#^qvOSXRjRgpXg52oO1+n;plo8|g%M38@ihsanew*^91CkpT8oN3`{|HQqU*B#f z3q*JvUt#+&{-PjP()+=N&S7M)^|n3>FqNG0@hUQrGT;2J-$`7FKs{mV%oM_E6LYhi z!4Ybsa^O|1Nl_omoc*I0(&u74Wp8PMD%`8Vcv`^ZW{=B(OM}BrHz{LENvis_1Z*JI zDx0#xiocwSnk}A+r#tuF(s4=3b7}wV>tk(BEE&`N-s|IxLG6}~sq4T+)j>S{j5P#X zL-`7{vp{US4{m_kt?)pph4fZ`+6UhJbdUwa*|8YRTWEPYg(R|{q22F+zy#bT<_&u@;nc!czRRTx-JTYwa-;nv7*-@bcDqh*%d|G6 z#@btKH9q+8y(S0Bb6yZI%Zo$XeU6A1j?#3J3IA;*jSFPsyTj4&5+r5B7X~fpGf;xZ z0{6b&v<2okdVJjpw8?C`$T~nVuz&{X*)Ih9I~QHOTI%dDN2+#S@$IfEJDQdGNXXhE zdkmmX^ZIPQp7ASk;&qnyX&Ccmmida$cRNr5Bs8D(rhX2 z>vy7WsMoOjOp0$k68>Jme&yx$0$kF|>E-Pp)=Fm(C1gJ#4)e^o)BmNR&Euk=!OCro z*tm;=l?wEF>_!pk-$U4-!PSWblGxF~%pe`a#xR*kp1|AmLtQ}H zkT0Xj2D@8zw%><^RFY>B(FS}Meafgq0AxSN(Tq6dI7KIysJlqf5oW-H7|5d9qLudE zxCb?oM%iY_UIm%)dOr;HqNcPSAE0fI<4T6VpAU9^>SV!C@I0qTEXh5S6v7@FFm70q zJZw|PW*?74(m%TNg<~P9JY@Ga~ZhS#*u(GFJRbkgpEQXEfxS1tWXT+tQ^OZ zbu;69;9^@yw~LnlZ|3<6K22egSQ^(BC#*c7kOxk96i>ej&lOmFr(h%dho95WG@y5f zE9DIuc@zV?7BY{BM<$lpFzT+9MJ7rl{|o7nD;4>DEj*6PK|-^7AGM&P8JBJ}uMIMJ z<)iLys9VAu_nuVPs}i13#6A+|7bAOW@*u|1MEHyQ`wS@ZDK+_TSCKXMT;GaRloKzq zSF4Z~IjSc7=|5o)gUXIs+x#Egoh{69b0Lch(YGQd46$evkdJY*!kCOM_ddPykC?zRCs#$ z*>E$lOu*fvpM{bVNM%Eqw8Q2|?F=aroz8Bmf8s}}qWh;P!$3Pm6f72U#w8|(xTp1} z>**HZ;4Y0S6`PY;j72D9{(vE$DKFFTawbvHsKMojc$wx6-zA)vp3xcaCBq%SkI++L zf&CAC?!F3>x?;BwFB@hvt{Bu1EIf)R0uwC}%CL2Qtk;KG3damv3DRNF@UF%Nt znP6&PO#C6#gqPiq=8($x>0H~P#04&>pl)gh1dr{q6JI;uoV*xPZz=xZ%GuVuO-oodAw@w|F|#++qk5RKo`<@Bu@gU9bOZ#zgIiQx|( ze8sw4-dpK<9vsk$QSgFS1N($aki7?uiNBX#MPh-WC_ZLxd7v{jLM5I{&HE%#!SNr>6MrJcFmD1zwG}df_DJVH2IgYrouZixBUiqk7eU&3I9|Ii1kXqn3Vt=vgJMo9S8e zcj~VW;7rkwH+I+YgnsUt^@@S-Zk#(R;21c~7GeMft^aN?TF2lp`us(^Vu>iJ0|oFt zWr zU7}F-qs=&LFyim;uRz~{HY!+;a*=o)SUw_Cg(&ejOfhf52v=)2gIe&y^pgbvhp|Y+ z$1^0Q8mJ!@bZV}=4)>}o#iho)sz|})%l&DgB?q0Olo}E?S8fz!E{7GTGr0aR+qk^v zV7t}oZOh|+d0_p*@64p_|NB_~p&LE{5b(_IX4(ZO&dL<;go|*V8EW?k-z9cn+wJ62 z^tAc>fcDC#o}tpNl8uF^X-`?a52x5KATTu@v{&9 zRYe6!AtR;fBC)}tr!@((^hx`tSKsNkZ){34;n&mdpDE&d2Xh?(_Fp!!z~7X#3w5Tt znDW|Q>VLvT8DZIEw36D2YMkuIFLMd{vjOva>E6Hu9vs;3?f`|`*`h=T#(CsXp9}!i z)9p@6kkF&%n}OgWB#H5~xubwoa~J2pUBqLj`Bd>F>bv;Z)OwIx ze&UaySjJRrEKJFnnkfEnSNP8y`+poL4L+)D_~!JmO{h9hws(j&@SGW58&wcHyPx)p zY=BJOYC;c5#i&>ID>%ynZVI|vK-&{0rqQ2s``S`U9u+eaN!-@s6IMn)eeRJh{)uO? z2!v>F5h)w&TfrhyJ@U;&Ix8-dufh_7v5k6c2?Y7a7vRNPBj!+jix8Q;EHaqimxzgC z+GwSX+Wn^Gr-vF5E_`vPbPyvM5cT!S&Yd%9g4xzdcN-jxHcnUc#7pl)^7Zb3B^-VD z@h|W8BEkfvB=}CnES$>>F@CECR7K|>fvd!z`hBYnmlV({{tU%UDCW}?1mCv;y|a&B z(7sP7(CS40xbi)CCbTh*oPPcKfT@#Z2Jp88P*jbpyvps^n7T4NWbp6g(zF?F=LB`~ z`KN2n{*HOM8y_n*_R31#)Z#_Y_8aKU)6oX9?BDqqWlQF!K2#c9IkK8ss~iP@@&3*a z{>Ksj*FWf@{_+sasUm{iRRv{OzyE@p?wl#7do1Pq%BhI<;c<)j@cI5J5Ip9 z-%6KN^c#o2t6{KfK*-MQO#$H6ib3~3oGh#+e!@gBv$^9BU_C#jka!&N3(*dG6)=@6 zr@5(~cRkdzTL?Aox0-lol2TN?0?_p2^I%b+Vto#puOx{JgNu4+%T0CXRFW}lGY)@3 z&3A-Y+$y`4qsiJK5#hiB2zE6f8JEyy=LK&|2Bz*69&m&VTCLR3r!$V5c=(!cp8K#x z=Y!MWfrv)s2f7Cp;w|TF&&%}he|+N9M`g%=*o#}Q`v z7JV?9-|4LjpXDnTPicFxD<{5bEW?a-`V&NX`h@~0?|%}^8#SXv*?xtX=gOoV4-yU9 ztCW~CY?4Mc^#L|Wz846<*#_lj_ot>BttK=7J#s@)w za1rYI(T__nt-Mqb^Oz<@#S@SHxILx?VB4o`=>7Qj<-*uelAJx zto=DBlQjdGvjWp_>(%imTUUc9Sk%uX*O` zCHPY=;^cQkcJ8No5^C*U4jhvHSfTV+j0S$4rQ<{^ z#mp5=!rN7(CN;Z6cqMpk*~hF4Q)l^eRS-cOHU^Fez3&6BT2OOe_~YV=aKh@6l70<4 zKpY*+18@x2iC9v9x5fVFI);mor0`d1vJeu1V(&be*P_!EOyqk(3tLc$kbWJy4!)9~ z9k-b7NXypYp~jxp9);$F%a`B2y3}t}S^s(%5hCYwBdR2VTH2K&FGvDLrL!_G6y&zM z)$gMPj!n+7Y@Vw1E>PD@^Sd>2<{$aKwKMUBzyGd&1LdFerir>wuTxq-`lmlT^4d)F z568rP7PeiV@&t$c$flHUIj|+_)`-V}<1OL0XQVs;EZh}dDsI%s?B;YCVy91yR+6ZB z;=NNF<_lFsm0PyYT*3|MOLF_bi?>d|qKb_3k~KB|i>hg1TBijyDt>pu@|bohwRS#- zCk|eS(x;?`650xPE$b^_C?$kj;GawyfK*{-xM)_rqOkTg#vqw#LOC!?H1m& z?7HBK9GEGicKJa!wZlKqP?Ko*Tj}Aw`3eRh)n!u4%<+qbh$?l|QA2XL<+M;q>I!@1 z-6!M!lrN+i2vEgfu$a3N8EVBVncQ9OcC=i}D6LC|I*$&_ItP4KTtTA{McHPgC(Uu4 z80k+7*(HL(CmRF@Y~jNR__6=?0$^(NTZd`?xT5;dzt)y8-`-gg3R+r(3F2M$H#m?q z5wZTkmxF0clRC`h(sqhM!g>s*S;L3zCE~`kL-DX~+FX^Y7E~e^Q&>>docAGj-ae7_Z214Jdh`=Mk`peW&kKaxW&5wx^e=Q6zVL)*+hn#bd+ zd&D*T=>N_X&Pc_7T0o9n03$`AwYf3#d`T)WLT3Vq6ts5e*X}n9Auy`Fq&Ju#xyWwrV7nd$lz;sJCm- z63ILvw~Dkaz{=@SS1t>lQY;t$Q&e6=TmRf7$IT{Yb=7^GJzy1ae#MG<0N*{qBIEM5G+*Dxth^BOC6*9qk? z$m+Q?$W9Gv4y#4sAtNda{|u&jnW_Q86u_JyYZoRCc2S+u!5`0xqCum|d5CgZ-2U9$ zw@Gm^3&SiQraoix55Aa5J{&gN@k%D>QCDTsOMw2|<-KWI1-_-q|9rM!274sy2Ea6J zi3xJzRxVZVf3<9v_IwK*LJw}|9qb?=Jml0;h5_LuXKil(}e2YUgs{pQ0S*ASqP%14q7LFU*Sx zpImHL{sLE+r$oaFPj!NFE~c==f>)~|tDDo6wIi3&(oA-&(=y$>Z9NTM9V53t(nQYM za-(Jp<^_qb-KXI5*XIZvL3$u*gF?UDxQg#n!g6JfyW%Ex3o`nb&WB5|9bxwp?L8zF z(3K!;jaDB}_p1%*7%ipKO)gBMPyp#a13uhUK=0Ifv=r|woYT{vZMDs~{b%W7h&95o z9X-9DBKJrq3@oc zKI>2|>X}te^fPPX!RniZ>Ec2)^0u3`H+!LA+cfBKZXm+%%8|CHIQ91aJFuC9KEofu z<3Q{~0+Udzdur^PWF}twJ}xIOVEv&IcMXb;6Mp9lYAp)#ZNDiYHcwk3lR?-H+Qb5t zQWQ4)ZO{d#5nlp1rqsd~0VoLFhE*sQvjzJbs3+%L+k3tvxh)ZR%dam)5$)sI-JmK=2ZuctEkXuu6A8PP36^2eqD4^cx5oG$k^WWQAwWWPw2x3p)C z-ptBx{_zg>1&Gtn-t*!JSo9vl)-9F`y9RNwi& z)6Bl2sjB}?RHWT>9L;8C<53|TzQ21at)U-a|bju zb_5bzO_n&i*x@tEkfAFBN}?d>>h|PbiWN*Y5UN7h)yKk0fM~Fev3An~JN$=SIPzWZ z94kqn=%r@wCeWjZXAj01a)*ZPg$Nz!ZOv!bi60Uo?KVfNMQyERaY*-F9DY^!%v0Is z6FRNm$^5h|jvRED=SbC|BqR&L$VaN%AU^$A<~uG^2=!BQkmG zD~xT248?gAr4-g@E3e61x8fxl!r_fVh#S8x105g> zAYa2EbR#9A1+ybT=CDXemVjkI<)3S;u`u*UrdVY;gDm94KT)oO`(bLV+vyKCf|^|f z8%V=l0XhQ@oieE{dO^q=54c)1&_(WYG@nyw*!{cMp-P}o&G3>XqIfmRSN&P3D>UQ+ zGP3xTOE+%!0}F|s#_x&PC}y~@CvgBS>KOcXjC3JIDdw|uKn}d%A0H$a74nb`r z&$%PZM-9zvW_{W`#`Eqf>24G7r;HBca|VBK<F3! zeTOjAG49?_XKlZcv4R+Gw+R_xlU4;}{I729|Lr^9ru_93;2CR@La@-u_*wlO*C6#W z{iYyV9pTt%uR;LY)vZaD{`lFLnROhEfe}Y}9-NY_Q9U8;Rxc~>bUt z)kJa;D?KmLQ(FwRDBniX@rDfr7y)#lJxb^rUo)7fwWluJc$IC7M)0z6KVc(~gU>^9 zfw8^D2D+YKd?e{ffK!lCS7U#j9UGmhT0BYet*GRcHkla2AVzw#`b{&O;%b;Ld`O&p z1JXizzjix9c5tbYSQhjT#L{TPV3StpI`y*1;n>@-V2TVHb~Q)xna#JpzU|g4f{T z75+rg1kc+Xo(N+uR$3Sxc}I1#ZF}YMxdA9NY%9bR%ExFxi<=Etd&uo$&nbuO?d$QY zX==ys8{&QN2|Q!VMnfF_yM?k7(7dq->m)hYY8^Eji{S)xqcimp1G4)R3EWMVZwXH ze{(wmc@o{&0Uv%{qjS-wI2E68e`zRAe_?sK23kD&B)M=lkV)Ju39@6=aJ zEB(2@&TbJVG5b#(%i9F$59F{|-!=-Z;J35VTu#egHf`~LQo8}M z{mfcXU(oCH@6x0Sk=Tz;M?0?8LV<0`bKBVRvsUOT&|3K7sdXLe6PQ}y8Xq`e+B{tF zNzXG^C8SW$;;I8qr}&hoqxt=sl~xZa&k*H!E3RbO0R_=L0&UD+m}=77fn^{VtKA&7 z!y~s^a6aLki59D0?c$Ih+r!!R68X`!w8Bh$GLV*jV${*taC0Q3g1buhX#Tx&mu7$SNnAfSHj|%CDyw(<3Dq~-d-x^ z@? zjmQ6cPv-x)fP7dFiO0dvGEaPLtJtx>iv${1mY(>xtvrW@Bb>>C;&QVC2|A59!AGHAvBg9gYl9~ZU?&p7cxyc zmrAlO>%Sw!v&H6L7c*EXWv}%XiotE}7dJ z5lj6bzij-T%*ZW?(zUq5}73zrrq*cXiw&v>m3FD|cZZPR4#AxyxO;F3?#zS`Tn8seaF+zv3Be_}dvJHRf0A>~efQq? z{;FM6QEWzfukQZ(TixG6hwjA`{aNLyP|z}H>1+b{VKqZ~G5&Z2i@Gqebt`?oDMMt4 zaL9fNS@Fm4sNYdmb>9OwUk>a9QeqAoY{J3Ah3`!t&*KYlUi!V2cv<8a4>mCi24{as zafl~|oYNKR+Yge5XYR~2+lD(%86IDKKoG7}pVQ)HJ3_%On!K?KTx^e)G3vg380zp6 zQ-F29{nv#Sfw=R)mUCQu;f2 z;s89p6wZMBe9%ErCejA@CNT~BB-6~_KhI>ofF~VX4fEG|`xEa?4A#Qp@A#8hW44#w^^Q#7eN;3a@km zdy{e{ienD!R8@}h)r}D|U0OFg+i0;}1Tjv|qizvOhyg^klfP5^|E@t{0plYPIP`^| zS>C?G#_Wl00~0cUL>;z5$Yz0hPjG&%vdcZM(R&J<46O^FL*h)u(wGI^WtLTqOox^- z0UKz>fgY~5-&M1J>U>+8h6rA56YP7NPn$l>q*56I}ypKMs}WaZe0!g8RQRd^a|E+ z13T=40&TR*h_GT=tNyk&9IKMyWMVdl5FiOcALWoU+tjFU+<+vsoGsU@^Dz;t3@8&+_>lV9V_bjgiE z(XGC+u2Qvzca2zWp9PMIZ$b3_47vPADE{-yGEV5TA57gRr}Qg#7q%}v#ZXnIf9+0% zPiX^P&f%k6?^ouWkc{QWWyTB?>07$C-r0z$x}T8+p$6w2t`HcTD#+aV=<605N@BvR@)3{+wqJ zTuM$45(7?=AI}{{B){Z^=;eCt(4ft$W+wlly+Wb(ed}`+XdB}?*JR9s9Z7>-32tE& z9IoW@)03q9QA+5R6#Hk6_oNyp*|MAX(W;O=qR8dRH1gC1I_7X~0jKti!8ldD= zskcL~rq7$eOkS?N%yZ(Adtuk*a7!eEZ7iJ)(So$qOHYm-Z|ruRef_2qct&eKL{=$U zT{)3;v|i2^s+p^IM?P=v_4JYS`{jPe|$Hj9M#X~O#St8+jJ~~)&~`XmQz`D zOYiEboRy%Y9zJ$KY8aa7Gq|gZ{_MEMUHCrr2Ms1$^hg%CNATyEWTeS6(;a=t=MdHq@~MvI}4*wb?{dd@--GG zZw7?m)Yi4P5cQhx2)uo7<&P4>iGwPAleMZgk5%lZp3oD>91W+}I?ebKwc9Jy*7I2fG!NvjKGjA_ySvLYIs?Fo7F?Dsr zJ5!-)2of7{yS9E5ZEyXFM$zDpP^(@Bv{r&RdM4bfJVUzc; zVvM_8QGk6^&TW8bSfbjkZnU-t=JyK7)mtM|08rlTaQHZD@x>_TZ`$wZy$ z<$d~WY0`bKDp#a&_@#VPvXSJ{5kaXQ#s)o!A@CCiVpDvfnJ|lrIg8TVd`QOTahtu# z4L65T$~|$_Q<-5@wfBIH{pw4fEUcuOhd2?H^0BN8DZA4Pxj#A^|2s>fH~?_1t~(nd z#%#+n2q=4)MT!PiD>`LrZuA}m{XFU!TB=gCrFIM648Zg=@TfZ6OsZv5OfzVGRSDn_JZwW(xG@e^N#0dqA znx5Cpn*H*ZmO>+WJ4s~~-69k-(to`cw8OY2IWv#%>yLMW2^KO3BMJ|>KP(1g;JT(= z66SBrqnsU^P;Do@5XiozBun4eW}2~7NiU@yMf7Z*JOlG;^&=3PRU z(jBGk6wlA5M{h-dRXJN)_lC^|=4by$lBAwMknQg7woSQgxzQ?ru4|^fPi#E* z_jw=hE}vMR1@{R8|HiLN@%$pBPry8-5wYH;L->{7x20rKO23MeVD~6qGCawimE@`>N=Gm%IQR&&*RZK$P|`a1W%PeTf_QqTWGRxqg+MTjH%ZKT~v)ZtUR?_gJGx+Hxsu5j|E6)|dO8!;^l-$vDA^(nc&-6Qv>Ou3Kr*mQwzFGsM3~ zD%2m5(p27TT+3ed2ulPTv=t8!*Xp#|El{vTu={=2-u>?|{U=ARI00~BG~kNrYwI}0 zEV#Sjl9#vr_>oN>qe$tXNoZnwl*)%iC*zDK`iB74Sh<{_@Wf222_Vws#boL)BWyXx z(B)YJwYwhj(yk%FLz@|OWZ!+0VG6CW9Kc4qMdIE_jGiiFf&0Qz;Jp8y|3b*Ppuh<( zEugd_jtM6GX&yJtT5=XtF6#0fH_d7Oy}y!A+ zxkh#-L=ie?(BX#nU3!eET6B67LgRSanKT^`pzu4caH&M=qLLZS(;xdyQ*17Nsz3;WL`Q0hcY_qd~wJ~X;TDH~ZG>1r6lUaWR zN|PvRUfx|(s1xcxutmeB&1O8d&PE+Jo+k-aSa}2_^g*3!vb(j%g1YgLK8@%AuIzeH zvu8p!DGQV_49Lxv##x_>0w^)jK7rVxL%q*+N0wi2Ws1#>4;rk&`O1!FJ&AslPrr|VZIbF z@xMQfhC6d+;AMdgt)U4P5?%__Hy8eC;{cEylfgFbB$^d(z6g%wm2t6cD?T13X204( zKnoIfT*y%K6MnU&5HkF`v#G{<^)!~&Zi@Pb?$?K224+q)zdG`5>ld`^ws@l#+oZve zbI)yg&NTKAX;kp+y2GX}nG?6*IGn3qTzw)lPhDtyCCH6i8pWUup63ZGrE#|DmBmw( zXUXWdw-nTfDz5AXL@Cn!+0hx3XrVPOdAir#-BuVn26Wk`*ZXAQ-eD>=_(ii5VvD)_nhICO} zH-sTurnyFA{~K`s4(1ZuxS?J@m}DP&UajD!v1=nmLQ2%5KNB^j8j!<^GFwQpuX^%HM)+o*+cQFv)ATF4#ztU7YxMaW!zPw}V) zt~K2W9-}k~S}?J=&D(U?E(`~_`3lq{H)?SZ;EK6l{Q+}L`lAFjsxz|=AN{k|vD>nk zc<{bj<>*DKq(B%fTCM~F?15qyFOMHNFSpEsI!!VAp0z~!H_G?52F$;274l5Heto^N zKk?LnvJ44aLi1BDn~qlKwjM86-mDSkO)=cTAcaE~~wT2VFPkEI@ZuYa-9>(WJObUz;gED z@MN&$U*YoR;tH?-q}%NyFtl_Kj-j>DxeFuS1$kTI z2D;5tmlFy~Nz&Cre2$FvF-gjvDEopkrCl&7Gm1gx@$LHAnzG*rV1){>_k@_H5|*)| zo)IinfyMMAJOx-{FbbpDR@7kpJfzjJ2`@9&YOPZ?-!L)3kI^|1gNpG1)$c3Db;w~~ zCJGtLnST?%_k6sklUi?A`Tk<5b~O!y?($M=X(=pUBTgtE;M>mR!7S>g zQN4cx#(Wn4o($dpg+E|LIwQE~rJvco)?uk2r#YHQ09m$ys%jVtc$@xs>4}_eU9nN}^G|@X2|vju0qOdq!qW~s!QghljWw(HF*^sXE@C7T z*)g}a$e#BHWF$3FI|LbSUyrhcOH6P%BKoiHOwrr^UtLnEXIpWg%=a*ss|L?nAi4v0 zTv8$TIH%GUF-q_MM=pI)q%Oh8x|T|*UtGP}sWZ!AwKRjZ38DrQl^AQ;BnP6N-w|#r z1+o?!q3B2(jY?{2d{QHcM<`tQysg(pZ2d^#3}SREvp8jgvTc;R99&S{1N`K>MPwU= zxf~szIE_5Lolaxp2ww0O7C>@&f0Q_ZvkRPx zOD>%5llK^ZbC*=HOOQs1cc9PFC`s)CiyMoo+~9AQFbPh0p4_e0LtfA?UE1W3X3<=0 zkL%QyEk5$2L)=Z5A)D3CJ6_niZ-d@Bu=7M(qexfp%XACyiHsWYSV}Zr?#R#pbHLEz zff|9(j)m$s&01Wev?aL@qQ=L0=tH|O@(y~FbZy@hw66z8ub?2Ad-oHsi3>SmVCcs-&Fhmf4vt!#H^+w_^|IOQ!_kQCyX; z_g4imb)!LE9yHobX`bcREt*3|tUj1$+bnYZZXy7olDA8Dt8s&N$PCX`mR%{(oEBE! z`+TCKbHN%VQgQj9qwEUeze zeLhVI5*unsrZpyr$+G8JpeC)Qn&Ue7DqTC*ck@%sd<=*!fiED;!kMmmK+m zE5g<}zAIvlDe7LMDWSDpgSK0?rdts50K4qS-mMPs4iATruI&1WiTOt_O z0>=v1CFF;pP~aWd7V??t6vY{3qI-Gfij9pVE!r2}D=l*Cx8NVFYDM>me>JWKJsdx* zhhW;w;P6*)yc?s?_fLky8yV?u|i7XN>meI=76CSXGDa$g@`sJxCs zR2^XrNNl;rd66mdl^#=4he6X9wo3gJR-Ws^xffU?=UOM`DrV+r9)8pJewUD2!h|EV zi}veogqrbu+C5Q)M5nC z3mZ}nySpr%lt|yXytr3Ia*5fh=Uh{8yj&tMTFH3~B_agD#`5+6i7wN}#S5a3dD1Xo z%RnMFbq_Gfl}I$WNeM81Swap=KfVBaoh!aYrobgz;|f%j(-=l`b*yUE-vfzWI1S#& zUXFlcbdp;yS{Y`8`FYJgUuOksbw#ykjG?dEPC_khs*GzMm-anJ$V>8+TmeVdz>yL> z=<1q`WDNA;V7|KC>%HZV^&8DFnFL%_6HbzLqC6SA?t#pYjSC`~U&s7klp1zP6=!Im7^%yaSV9yyrJZ&+-l^IhT*(ua0LuO`o7!7{+KV z!mz!Qp>(Y4N6nQuk@uBK`aeic6#8VyVE4<_KflWWH+>B{4NOQu> zS%fd(Ok^#Ddwmvb#i4?OU6hb4f&!sH4e4Y{akJC)uuw5YrjMXz*hG-S^-R|R^Pu4?PrEpm z{}1pl!om4bh@1t)M$x8nXx-&ygn#iN5^YYDh?Od%MheAC)r86$n^hb`K(1Sj)#f$e z`U_(0ObBI=#@0!pV>e@(^KJJT_nzX^Q?+nR zq@S^#aF?IDy5$BN)|CCRg~YJ@=Hx5S_%05zBqEN6cPI_M=JP$!y7-r7w-t2(^}kp*ou1xOmw->eznKDJ+UZ$dfnJ(!liHPWaoC&Fu&fg zl?8!*6z`=DDE95~dRy|A(&Us);t?a0oWW8m7zCS`cocMv-AuKT%ugM|qFM0-%*P;f ziNDR-yhNdX<>#IWd%!ryoM5)~jLcTsPmn9`C%m}1|4#hJA3@K=Y^)b^6D#+-ERH3i zj=1|IT!2`@M^fjnWJEx(L{CWA6hvUCf1JKgDGs;b*Dv?%3v7$d+V2-~-|~f_NVs6U zwz|&1d44N%IxMWpUrw`1G_Ux{8P`T6v0?;&4-13#&aF0Hm?eMOc3&!(agK^Atr=?p zyohVd`W_3Rll!&&Gs7^m={THF9wrg{d_V#I3D3(;=!Hk^gPTX!2A8)xZatfuz~wJbb-L3RfnBTiWLFxQXxOe6-2E2!EPaA z1y);P-zuMST9`1K@g(a7FammlE#4W2E%qWq>~AS+1x_>?oMNEB=M?yKm^t}PCX zoWcA*_@5tm-T3YRUT$V_{(SK3#j~o2$U4ultT$Zrp6ovfIPGP*{#`NgujnBofssX^ zQBZ-br&5dU3SCv{x6meMXQjb01A4tS?R?t{rReXW$0$!bA&~eAFkaSJ zrA(1q>uHsuE{mPqt)c5?>>Kz?@(W^k*su8nf=&P{Z(|v5HZ^Qz3Lvfp5 z%o0!|smPNDj~YBfcRV1Zu*Yb2d$e9ZY#Ih4IVgyWuB^axb3$8S9yK{*l(`EvxAN;L zQ0rlY^s34zF0c-*#3}^xtyLo@eS!U`v7YKBXSQVA>NLFn zocwtbGeJa3Oxlt>UG}%{N5r_x35h>_m#Uw?SfkmxYh-c;@Z5tBy~~#3(9Um&0m@Trp;*)BXH5fy6gK5Eb{u-$L{u_Nz?0tnI`cX-2c)=C$dBwMkG=BLLBgpTLUFWa&m`tX}Kg3;+7uHPrC z5Lqvh@^1Zf*Zm|Mq-IhQD&~~A-m>@;uWu{pP{Oor#T|=9@v>;eW>5r|YX~vAw~397 zae9%s0-n;5gks0oFe>p*9t3oel?bfgL5Ce7$=<%~qNQXV!>a$9R(KM=B3S`PdlHGq z;Uz~uYVXmUWiT4eS|{oqxVtB}`uf-DF>;*fGe;RZ_y>;bb(`D0M@&_U>rC}Fc_hlIBB3xfCjmV_9PKzHh!Q-K*0fb`! zXM9R%$sEue2WW{X=o`WK7t&WJeiBNkm$@HKJ$N~;;2R|;N$SYB0DI4yF$#%<2f1RG zCT_N<)*a?}EFR2$(BrTJ%DZ(cCZi8H_ON+ax3)L3AlAxG2qOJ8VmPt zmxMiT_Ho~jv-08&$5abStob@Lm5V% zQAUrVB_jC2;`MXMKJ8@AS$VQ2T^gT9de&E`>WD##c<^il9&Y^v_J?C#4&xgzP2pmu zEI0jVL`f-$To;3zq`f*#xQohA*)zt*=RK`)uO#TUcYd4(&x;UK4SH)ls!Cr=!xjcL zyfNx$LAGB^H->GJNj9bMDjQ_1dy8mZ8bm7SkO;-X*BPIxY&`D}uvb>@j>vA+y~pSY znza+!dGJEb0KU-(8yibwC*BIJV^G^h?Cdx=)b-C$EL^ZQxqYBtTnjYn0%S+uVPS`B zlpJRbSxQ0cbH{@e6uG^#l6$FG$<22T`42ml{kBsCcbgh*haG7F6@$-5HYHU!4-6#5 z-bg}HRA7)q+F&TDn9@6uA>i29kiK&4tOlo%T`DeNPcr3Z?Dj>dM1@$l88;Qu+%(s1 z>q!k6^5Cqs2<;hWE~idrE!Ww@`B=VVda0N>5}JELm3B=PgR=-?!yR{+M}5CLak64U zPxppl^6JttLzQ2$4BWmAc4cD7@sEl%!K(z~0od2OmZdh#q1hLDr+bcg4~9U$s3%SA zuF-R4TzIw=s83J0l8U~HiNmmrefm1KH``X#erKknw;qoO`Wg&Jcw%cY&Gq$X)?Hgh zskNxj6Pc4qNSKZ>;yvqnVVndH{xe}U+FG@j(Z`cy;KF&*hx1q%y=$%45IoOr;~MiB zsfmwhHCFk0Y|EXWKXeeKW>E~l$cga0lpZoS2bZR2lofYp`!kbJcW>yK%KQ36MDhP{ zZt5h|*VpSCv(v~w@t=UO(Dqs0J;3?P0`yE9A*OF%FK}U9&?(H!4gL&|MQgc$|QNde6QDJc%0HzsSIFh>DD zyjt0Eht?i7I2C7$XiuS%rfEt{({-sTO$@YzJsy>-Iyb)0QRYeNeH(pgL1v<1CC6F+ zA0`0-&7xrOPEcfReq$qWxDsT%{%JB%GfuajQvoM-K3(tKK?Cc6Km7_Od>6zQGRw2a>&sj*^}FCPLnceg;_7?HxD6s3n9XS z`jTB;?P24XSpKNQ<)}PNk!br25&B+?!a|IrZgr6 zXv5WMLhw>#9JM9&a4OR=N?m!6KOzNuUG4a8=e$gLX@_-qCtV+==Lw*f__qru0?6!l zdQuj68?)dSu-g`oWJ&)Ogpw_!kg{xW81W5=D!>p+`>n9TfBNA^{I46(`+A)YOX2cJ zVR@($U^sL}m2OiYt+J!Ar!B618u7B0?>RF2fzuKCWl~{BVF9UPDA5~WbY|Q+Jf7uX zhT$8`XDooCcZ3Ux>yIY5I?dQ#`W3{iJ}!j*<8|U0&jFc<&Un>LZ5`d!paKwN%!cq%-s4@r zLo`XRn3_kUvO#T`3}#t;e8g_BNDyqjazZ|Ap75h;{SPZh07lh79Zce%*D~b5XhEWt zFkL8+m_b+}px@P+Jypi!))w{PKGXM^A%kR}4z6h8;87d9QY?rcsKK1`0au@lb|;7D zgQBVmuM{CeE67O6!J#Nr2HDVZ5B1Pz?q`OM=J(b)hH7OJo|#tRh75DN06OR8EhsW= z_DLt~FaQ9vngC+X;H;Yj-OtuDWR{ znX!MI9z^wHqaTm6vb1HihYL`BN|SFWlaU8fMs5Wr>s`OOzW2J9(^Z8GC}#vCMsGtt~%FkCtXR zy$``k?(T{t+tKE?Tg1S4ITik;?Yi${2*^1LRi=&u)e_fu0&a?9SboTU_w~N7h7gi< zX_14u5Y4aS&jW|+$N@qvnA?SgTJ+e*v3Y{q{)n$DVhiP}$^ZVuok-v(N^c-?cMLy9 zUB^r{^UaJok~jrD&JK`ARLe5edhMCR7R%^O>k`YDT+d?&9T-xabT@#CzIat(h?(7( zw>pJVTf7CLnugoq8Zcjaw=!IQ*~1@SG&3McPB`Bt_hmk<39rFBzWPC|oTC^%%6FUQ zuSfCE&rlBlff(WnhMfkb3UaONyX=NKd)!E@C^3bqsme-}r%UD$<=mR4enj<4^jP68eKj< zjIS5KK>sUrPLttS8b`8<+l^#qmvm-b8(cM$iclrqgJ12whS)Ndjf-h;5Ay)b{-);x ztxNbL2^H!LP2+ho&@#+Ss>&S(>KR#TOBg^eoNrSc(Pay4^`g8bpb43*^;5oU0 z=DZ75MLEK@5mr+ztg?%>Ugt;hia}&I+-M{ALRIqE!8_LIt;{cAvXQO|9XSJ z{O}KIWPxt!{~%~*`VvJ2ArZ<R`SV-$G! z2kpb-Ywp7TWxE-Za&~w&=uC&AxDdK=rHAVI&DKC5Nhs zg6-C4$)bSwGAm$C@D*`m#X+erNT00n`H$8m+>tPDPycr!@6J_{Fwnb2&wmFbkRTXQ zUpj>qe{}kYUo5LBhxAw%JKN9n%zTI8ANNDS{_TD{B1Le!X`XN@xoc52*K3w|Lxu5q zow+hh;BmY+>*Il-Ok;08G5^^+WnWZJ4w?#ACLa$Br8wAQl7)#R-7V=zh229RzsR=p zXEF$b-qA(L<+VNMcD0l>FWT>&Ha1t9u5e5rh8>l+%MLl_sVQPOz6yk9WXbQ!5ErA7 zBQF;SjkK~jyCP)ki1oBCJi2FuCatfYFX0QCzbyIYXT|54K}2iCd&c~C%^HdcP;yCC z%dn~PJm*DPh^==~t%| z`@t=>4&jt~x7#IdV-#FD+BamEpA0L#=TNNck~W>-nxnn>zUs!n9njwjNB!e))b}W_ zuL;&&b6DKk5_*v;hmf=mPg>qbJ-S_W;pyg!IVFe|5-l-IA0RN^qxxMFJFiFHwQVN8 z9%|KxSF1+|-*h+5WRQna0eLPo(yO7mjUIw?CGdE-g^{__%?B*;3KawNcjj}GaPLHI zVR?V*7Do2CdS&2jd;#fLfXd#%O0`FLp0}w?Nz7nqMMC>lx-^{o z4;hI@$xL!Wf6H^z^Ir$hn#m65q^q@YH)xvTu?NgU^zi=WAs{ESMXOEjs~6Gm%2(UJ%ZiKA$KnHo;!Y~m;2Sqy)}~@39N48E=8UTWoBS;l zk`Hwj-*9(Y|Dy%4xwS$1dT?~W;r9dTWvp^53{B~VmlO5khxjMRjFdo5f2G1fiO{G2 zhLF_jFwBHD$}&P`7UlKA_W28Sv5ZA(VGlLqM;l<+d^((ijBX~BN;^8V!{;tR+Vj>U znBo57eke1;lThR$zL$vAi&c}lETreB3<*3Tn+65IJ~@>$dP=L4{M!Sy>f4hn^O|qk zj&Ml5%f8G*vjfyUVu{{|f(ib8>o!g1uN>tt!7!M@zok1i<8x=yiDugGAOBP9qevru zS?2cL0_W)GcaCrYM(Fpn?XFm#&mAL$;M+BMSC*7GcBoJLJZb*5GN7+{yrsEYd@ddKpTEOy zkr2SiK4!uA*^$-?ow{ZyI(n-P+Uh4ia=2ee$DzpG5qRWJ_(%>KI-(M3(p(A zh4R{w(3y7fYznz9*?HCh6n`Ts{)s^WY7EOpaJK!J*||{%01?K~o_o{EC!}*v`CR^| z_t}=3ffJcOgt)k>=$F2s+Xfjn$LiGCW0Czbg-P+|84QTW9UaKE_02NJZ>W`d6Rw2s zo?boAO4dR`qu(m4a28NRjE|^Ouwc^Nv`;8XY8J)661r^ld9T1RgRxEX=DDEobR5<{ zr3K8KmoXF(6fpxaj>v3}D?vp(rXPZm)9GK#^yxvHS-#;a3EzYOPmXl12WPO zTmXa^x|H@4MSl*Pxg)`2DaYJf9KYexP^puswZN@OURu#X219c!Fe(y~KG~? zx>kKL6qZ82dEQD;OU%J_y+aL_XL#Yy>%8VW+p0!|*?H3Qjt{3|BNQZy+V@m*^F?I< z;Tt94s39e+!bkc94L6j?osM*WT0n1TsvA|qhAp*g+!3ZSWHmi5Y@TwQYb&WA0f?D* zFVlm>f_w{0Py8c|&W@fsMDhzRwvE^ITnQ_AwaD-P0o<`LT7t!5+mK(x_WAK3kM|AB z=!X_w|Il%1$C`^CE#WZWWXspQafS0bqNJ-NkgSJZ6=F)$*w_77H~FO*!byuHu=MCC zvyrmtHl|<_CO!P*1v$~1chzuV$6`7d3rzemT6Q`_I*04DY!BYeu7mh>+f#lO(%&oT z_%pN-c6fA7-aYh1b-_*!G#WCC^7k@R2Z)L#7Ck&Y^R)3fezaqvR0Edr0rprGa@$x4N0ZBv-eGqp1GVE#?aDp2f-?>2x`hEiM$E z@4oTF9W3)Lmmaya1;TCXO#6xdPc`-(M&znH3v4wc;-7Y$>}|xW$~hKY$eGAu8S!kY z!aXl#-*J=H$=%645BV&soqePF8?#KiLnC>iF<-aq_zSq9Jo#$Fn@}0c7QuK+1!^Ib z?e?tF4aRX>(u<9t!>RHSKS%3%A{dSva#-UHcqBorY@2fsBnnjlq@k&`6So;7`aPE{ zlF-M*mEH05eY6+|N)w^=(i;6%@O?5gUOk`6mZL$^8QJjKhSN;iz?iH~vt?>Ige`@A$s{fcd5*lx?IkqE&_rT+{)}5)@@%DESienCR ziyt_kO2_uO;Ff3|+q?tpOj>n4^;sllaKXDyIu5jEsea<%0I)|fS3=AUTL-Z%mOtp$ zryeT=OL};vnEHmU})VBE>}56lk!U+ zrj=s0a{cd5%Hr<<1L#fd=|2-cN6Lm^(>-?rNtz-CZri^E0eNV%W%+$wShH!qY92%I zXh>TOC!n{O}v;bc90T zyHCi<%7EXIq)q-Sg67*m(n=n&f9B` zyjgGq5B&Rvc zV}I`cI|{G-PGth0AC~$zgP`?nAo%eIb@#8*!DI3mq=1H3x&uwH!!F1=b?P+2KAj9> znY2mI-x?-mP2b9{c_?h1^4-6<6(S zL*a?fylie1mIJsd61i+Pwi7pBx3}n2yPmmSdE5cr!(ivn@4 zs=w_DFKYMqt-(xQ`%}4Qp)TAB9jvc5JKDH)y zC-oBocmUb&ZvQQ&C&O0O_YI2JDITg5f}6Nn14lnA2RPL{q`cI!R3+BvlIJ^aC4FzH z1z#L|EYTt~u}>H&xCPuEJy|d&i0n;tl?|KF z^-SY?g{a!ZLvT`4FqQj1OG+F};eTuLlzi;>{K(g#YeSyxVUdk=Vx>QHUds z;fdBh3VQVF_(q`6ohy1~t&w1cAj-C5Acm!=`6mPH#MBx$sf~>s9JN`=VbAmEuq%y+4bPPO(!hFObbWR=fWlC_-gm zRgKK7?sRftV1LAT>K=HO_QM#NE9?k46t-03w;F_y{G#>9( z1eCcYieqh>^QPzM-+Y@oN`5^dext^oq$FWCQT7uPlx|VXDftqHq$NJ$^uh{2uVD=` zckN2qOrHx{3zPkK^my{gM~+d44XSO08VSuCu^Y{Hg@a~`?id1ItX{cIUz_}f2RvKD zGkFa>9dJb^f zI;9Pn0BR8 zQ+`nS^9u9n0lF!LdIW}be~9WXCl6e#;I6KRq2WqMNQdiZ{q;t}+Ao{_%~x*+QUT8jaZr92h#zd9O%{t2Jk0igV6L;NqcvardOszdw&TV1wf}^>j z_clVj5N#l`u=wu{RDC=_o~pOsdDD~r?)HTC*P94-X}~2=W1>Zcl9VETw*5AvVCT$D zkYO2IxZZpm>BqV2nftRuJ#ONTOJB9#sRgBots@_J6FqrU*>bv=uGR)Y@6TBtX4QMn zxIBHkXvkdepT6)nrwor5N(mrgVbMu4UV5HCfLR;FE9Cd=M8?Z}4-}jZ%SbO3C9;f- zMEV7;#mYBy_f%xt#YK-efRvy2+qy})6!Lvgm4LpjPvZ)y0}Lfq+yex-l00d!Y{X4$ z@AsTkZs28{#5KON$3xA*O~@j+nX~V5o4Y(`Sa0d3=&Jb+ta|b7Dr7$G9hMDmw}_oQ zPa!d8{TcM}?`#CZgGJNsM=aF{0qZ$dRBTq-?2TrQUvyDE>uNNw*=aKT>RY6fsFqky zf3GZ<&RT|=Xs~~-a^|}`=EF;JG z-(GExx6Or`@^wt$5&rU-9CdcImUGc5~ceZA}`$vfBxR})<=h^ zyB|Z`KF>$4gv~3{G;v<-s{bxd{~_SPzrl`4gRDk5DUD5VXtq|ba(3crqb>GvC3_FO z3vdX}T`@4Uz6_}*X#-7{hZG4`xlxCs$O8*CHA`t7=K1P_O0ItjGmm~Ngb`i;yy{;s zzp8AaUZ^U-=aBX#B9T#5`UIKs1q@EZk``EpF59NZ4x3-~S-BR*%(>r+O$#s8_Jf0m zFd_r%QMLBhdc*R?Q~2Wpv@^U>R=>1krjawYyO;H)sDnP8eZ?a@f zB%<(b7CmKrgicW#u{4`gh#sQB|5Pc^-&M+SBwbywAs(TkQ!Eo_B7uASt*naEbv?yT zEMD6a+3BYNLcn>?RTm*xi4DGFaC*WI_W$5*2It+%_Nv3T3T=mAUIitoicV>4#jhy; z&~bk%hmV}#3F z2Nh$4se}5OJeK{W3(v(}-9D3&xEPo)nvr8?R2=TBu9mBcR%1@&rT}Z>Pq^Cq-p&y~ zfQkguIB`I?H`{7v(@*bV!=usjoTv?-18Yqg=H^#p4&7TW3qStQ2Y+g_K-gFATb4G@ zZac>?>3L(a_vI`C<%b9kq%|#a0 zfDhYiR5Y<_518S1$(y$gzt{%%Qw8qVk)!_$u>HwO4fp~AHh6W*f$cW@sdKXg=o!C# zS9;~M$dmHE9r-aG2l##cvj^~L*(|yU14}}UIIEtm!IHPqt>5M0FvvT7Xn-d+^m0H( zAdyYc7;5KJ7JW1q)9RsMavbboz{`AqqkB_S<}a|Ll!y_ZSCW&N zV|Ow0N{yG&R*(xXd+((!X)lL(mK?QUT*XS%{Pvu=;TJt}f701XJhadI?Dq~(vZb4q z!Q;JkL!kJt)AjfINr-=8zQ2D`FQDd?>4yaFO}spK2l-|d#|Hiuth_n`#I<0bW>2?wTXn^7@NT+ecyNGx7Z_V>U7v>Sug^I20jqqnjb5L4%dYIKyP&B zX=Z$staW~e7DK^nOrXt!7o;2WXBR#X=h~d@cwJ-V=xO2V8X5Q2Q0_~X?`z7WhD!?A z7BnSM9ZP}P!0-_OjecIBQ4#P)71UA4EdHIG@=uV0Z5WUgs+y)syPS5GsHA@A;9p!< zc$4%Wn@HeL>Z+F4Guw8Y$@pl}5T7k?w{|UzIND4(aZYM!LHj0qJfo@qCc4zj>eMonaXN zxifO^KKrb__F8)br7_l^>6db~W4qQm1u^d`cfUI{{>#ieT8{q3oJ~kU=ic1{=utQ< z;bnnCx~{|g1_y~#5Sg;uXKRj+mFJC)W(hG=c70D>z`W~&aCbeAP^x&OFy$jP5;=qi zm3j(JHMoXUPds-$aAzhjVH(fZp9^&Y^!@Mu_triB0|OO-Gq~(w%Vqg>z2^3!i^45A z_`TGl8U^x-vP|pC`#ASs+(+&05D+-@ULVX}KaQ=mP^&6WqH}wSBcJ&VS~03@aOag? zlT4wdRQ|4uu!f`rU>vWad~%wD*aB^pxnGu9lgSqd4{0eKZ$}cho}?(o-9ne_a7LnI zRfD6vSJ3-m=}chnF1@EaTeWIo=x?8Ao3-@!yYwgo4wzU7B*Ky|o9(XhqJHCo*u=uu zdTrH*J4tIPZVxT#>m5LmeF#5Oc4pu8+_jvK80u#PEKSZ$MlXLUhU|1gDk9si^d!G> z*{~h$y^OsZx_=1Nc0Y)=zyRb6DL@%C>gC5H_=NCt5n})JejSL$$%QiSea9h_VPj>o zwDu^=__%9f}VGU&hP>) zaCp{i7rmnnPFB*Ibhro|iB|kD6GX;BJ6Wz|PFYWI<(>k7A%RR>)9TB4Ly>rep`(XG znePko3w@ZP7GfNmA7cGlX$0OU{jQADu3WwwpXrwO41eYR{v|3mP5SR9@fIY(oB76% z<&CXJyAUWdJbOh_;{ALvE5eLdM}@#GFXozvI?SrW$qmiwtt&$&O8A3C%UP}24o|wp z=Ti$UzawvB-lI99O;0c3q3I6mp-YFCs3Pz{w>e{#!T$MEpy*;t{G|$l7?$AQ2-uSK zp*y&`{rP3b@c#cw2LQNP?unS0Elv9o-05~Eyb&(M`48NXjX<>m30Rl8$Mjpurtr1D zdLxPTNf%X~jP)nv_iO{)BS5W7t-Gkw-`rVT!aC{IkeZ~5teja~2wDU06(G1S5&Ywn z-FV=cSU3)<`Dw{sW*!bsS#skNKvs^*qy14PR0GEt>hP@Wk4Cy|zWiB$RiuvzLagMs zYU-22>4eUHn=k?9gomY9Ea@Shs;r%9@No}PBj6x%IDsFGS5zRcV28v zMVC=Q@KjwO;uO5jF7KGWd!N(q#EFM^Qj%|LQ}swKXbw?!|NYu2S8yTiX`X(?E>a{?ZWv)Nta|a$T|3U1n*Bb5-vKm%V~>e%LcSp6<$9NSqh!{bf|S zD~qq30~f4Xv(ety%v^x=4o&h2e#3hZ*kmOWI5|LGucV79n38cscX9uU`*%7zS8 z^TzUg-9gB)PNQ~x?{WXXBecgt6F{NBQ@Mqj+z;bAm@OK-_1!(CY#(*4M z?-HVCw@h?L14SB$E2X(08*SsyKf)1+9wkf?-{zmfu) zmy`eJBv|2S@EKZxX4LacsS*Wve5)D#|J|QfhTr3#?emG|skk#A?9go&5iYGf864^k zEW7a4R2A3oVZwpUjD7ddV~)rhkhLiQ10^7xUGeu0{mjP!m~{-#8cwpLl^eb_mq@K! zF28))dtPNd6n1`|CwmAxt_tWxOQ>0PF&|+R84=1)ECC(e*H^!-(-k9ZxAG4?UGK&9 zw@Oo|9fYwUIK!LZYg@W)2B*r4!|OsHa^m*v+&yF<4uPC;|G78u=}`xL6YlDbQdh`8 zU{$2d$tN67IN)7N0m{e!59$X-1v~ZQ6_uzCvK0OFxE@jWVjKSp&hP*UHs;X=D>R~o zlG#5sDozAa13(e0ZXO#d7dTu0fE49#*=7r-vj}sjwLERZqqH9Olx;OurkI4|(Mj-gv)qm9PI$-Ler~aO&e_l42z-rNZbejJ#k@3x63{;4+l@kl2MNa{ z7+>4OA3m#6gF0mfq!I1!(PDx-dEDw`;C@h)tr7L>SyB z`?WLn{fyS(9~pEA(4D0Wc7fKmk;dOF`afI*3KjOta@3^;(Xd%K?!Ef`tdEfTZ&R`3 z6~tl*)=R@gXsdPuv#p(R-}8P+vxPKIU;$k0!j32=t=WKeof3EKJj@eXfknlQ)*VEE4&|hwkW#*u zU0$L@i-24bF_=a~e%sVdC+Qd?oz9P9&~Fgu`2P8+u0Jb%DmZ+L3U(Q2@~#sKkv!oh z3<$y$r&Z9nbX@&DI8e7q-mX!y!mU=k;$OGrsg@8yf&h!s$N>@a2b0}mE&qjT^Vfi& z=&?Mm!Zzi^Qm<(DicY)uE+Gx#N6X>~cFwR0!(N@F@WRpE4jy7Ul>BW9tI;=!_SA-YR zeEBvxY-q!$rB(jS^epjuNgg$=ipzNV{NP0AB*7zhb|z%43FQ8P*mbf8Te5k0?FD?ib^!F^9i{K+;kicv~f&{5pCMgDKh`o zliJuU?g|3uUg-{=)7{q<(<&W^q3Mbx-=X*6$goq z&`~VRG4A)%B)FeAG08Wi>ZW)!mqhf}9E3H59+(^~X)xyLJrrT1GtaHyL-zI_<<271 zz~c)#Z7iRD$|5|Jwr358W!Kb!=!ZbPE%^@E<@2}R?!bfKBoH?mlgdlC!CKHd{Jl7T zqhuXRJu&**K;_BCfj;h%2R0RH9GCjvL>%xI&X~D?qf3Lsh37T%5A~ z*-edrQMQPCVziF<9qWm`jqrstWeG=&DWbFsfqN=1mPF7~PYF1#$8<_b99^8x?a;0N zkkb)aSIDGfX~k!WX*KY6=d!Y1RhVSk`kGXpKg~++Yr}mX!-|IFs)>bF0ep*q@gtH$ zNb@N?tB3mEa1~o0iEpOi>0P0|Itd2r_@6)Kg~{DAxOZv3UGI%lL< zdvl}@4{$aEbt6wqB1Sd>=b|e`m`vKi=vLwmG3O%@1Ne1tR+Lb=?zgza$XW!SJLVak zjeq!;kumNOV%PhO=6dlT(XyQT)m}5iM@CZ zy~-7vJ0fv$>J1kR{G5@E_}tvABPN%C(UOJtDVvS2E>-{|obhze?k?;&>GK!Jpm=lGIbx+VH0418^BLRK$+VJ7&b~J*@>fM~O zte5sK;`u`fejHR70pdM?UgLv&%{7c2CcDhE)If}0S567*!hr`E(BGr)g zT8n3r@4tE;&G3CcxT?C4N^z=mH?(OAN23YObtr@p5WSAdttbe-hpns`rm#=#vfTOS5z6*Kz#m@~h2*Zxt4(ugFT$gsxr$j5j9ewVKzw_fw= zzw$97-_!MOUB;O_NTw%e74y-IrY_9-y%P?_I*>v%X<-}fSO#U8s*-biqFkyv2fVsG zeDKlC_5oy!G71vU7$DMOXs5(ooMFL7e9E^JZ8Mv33HG=3%qrdNEF_s>CyUYC9u*RHi-fckdE4SIT5w?Qe+Yf z+<1$s-6@?z4rw}0HBe4%$-&J})1^Ytg4{ilZt8|dw9)~*b>yKyqBp8X9nR0Y&!1Dm z^I=w}@n*dr&Zc=E(To zGWbzme@cuHtWgs2zE7GoNCM%TYeGVcmyoR_6&d-9k*%i0id!pi!cW%*4RpNaaVv=0 zq7cYV_Z`Zm=|o&d^AKNaFZa-_?Mu|F_bWH3Sa;T;p6Kv%us0{m%Qf3%M?L}QUdwy+ ze_mAkxHJS|RK)Mv|CrO#hp}ayY(6hN#I!S}aM!D{$?kW8zQZi@)CZB+ab8H=r`mJo z$U_Iex%!^MK>hnJ3s5rPt(JRi3bi)4VoUKXKPBJa=`xLe9{?GpM4QUfH-}L2y>YmY zO3peuHA&o%FlcKQLw>AG*ZA#mp6)IQhWdQbYguNE-!+EY5O@Xa_zq}Rgm4I^E$42x z*7M_&G98mrQG-+mo9@1of7g8*A#6C&HZTpo7%wtFL6?Mp-CYOS7<_{O#grfb6a!R1mb z?&QG52*3-jyn$3au!(6vRXftTxQ?ogp=+gJkp$y2DLR_0E>__oe!XNlbXRhjGeVBl_!XZk<2W zpjD9Gb$Oqsf{A;GJC@(x9I8JE@M+=)vRn@~g)JQ|Hyec9;w>2L(J0vwB#dD5Jc6!d zht5YH8@QfV68e%%2}s)z71vXlXb6H$g+8wccs6k z(z-e>Tn_1qus&ZqU;ic|Mt@Uu$1y#V(dlr8n=Wr{;@;lvJq~SkDcpmyNh#(E&UP)~ zTWJw&m=Ev+uY)vBTXL8j#kQ#kapEDunfBS;kb1b7JAK6R)@APIcWt4dByAxNvrKU| zFsElOGl#2pkwynYXbspo`qIhwCLr1YkG$@)5XsNy@iT*u;zM8)8juUSzy{1ux$suF zRcTe}oYY(7hehf6hv1?w2!MZPh<*xLty^*7QpIF6H122>aVR)Jnc=)DAY1!&`&xZQ zyXU{)bAvPEb1_iyMav;J^Z4`px4n2PA=UWO_+RPwJ$d>lz3M>?gOd~@P2%?)GJu=` zgpb}``da-5zJ*8=(}9=w%!Dh$KfJ8^cZpxav)kLMoE6U}-b_~5>lR9uSo$D;XaNmC{W z)n40(c+_l>#u{f3-*Hv1pgv3-7Q~`@!wIz5tH8Pb^rZ^5sSvRn+lu2|RaJU+C}=IY zz3b}uu1-dP8m>ow4(8g5)1aojvnt}3B(ACDEZ6XZ7lQ81Wb+HpgO<(g9_eKhAvEhw zdf*BhoDVoG{0*3|2>gIE2QVPpfloSsgNtDH@AGs6CyDn># ztE#KKZO4JJY!+lHAgp!oIZuRF-hs@wqb5`D;6^3%7d%$VZ9*wd{b`M20;IKq)c^d9 zk_3aX05mdDS?4XH+NFRnq-63_VPLXOkmdZowfc2IA_qZQ&hW`Z>d*IzxKb0T#;nL6 zIvs>=pDnPA-fz9-f^W(u8svf#mz8F(hIH}f2%%`}4Fh0EJkxtNhmV1&Htks&LP4{ZrT@5mod|WP=2!!DGGZ`efUQCdF%>@whYoA#+GUW8>1F0;rH||IoU>^L3438+wlrwmZa&icG)m%2D*7KQdj7^^+mz*k{ zIg5QKT1;e?tiPtc6%rL$ae#4jxU^TI5``ZSXn$xF8lS<8x7m+;QA5tzF4zy**>7J!a`)K@> zZ37ZjFKpl|IbXRSyi_PGS9OyhD|-|YGw4mp{;e7?s1e3ql6IXKbS_*_+eA_4RT>8V z?dt+k)o+c{ji2C-pvIe-BeSZ!(4xj~IL;;}LsIOt2|RlP3yTB1d|_VyJKhxxTfEoR zQ;p!Bc7-|$S>2Flm*$fxIP;VyN}o6~fGPd{L7vEjO& zx?9I9 zJf9?S)6lTz8Bw2&2|kEp&*BTKCwW&`5ZP8OU~5IZsnFo~C1=W{nve4LKrR;8?++g& z&IA_Tn`H`{!A?5Jqcjdo!tEwlFmI9Km*$CY+dTCsI6}I3r2q5%!l*| zhgL7lwE4t)YM#^jVjBA?q3GSUrZhrmgAJ`bYs;hL^BWs*?c1*R%50wE$&2E9y&Z+> zbkTNA^DU!R#*|0k;ztSyEmeADS&1j|l|#}gthr;(XRK_xQhc@oYN?J~e`>zc)5|5S z15IPc(f>V0ZQ+lSD&M!CELkHB35!3zr_d}!k{h&9+3!?=&k9@bjLTGG|w zNeS%jIP3h`OjcX{po8A%T3p=Sa6WVhzg)^n5+x)_y(R!t`ct%}9fQN*s^U8u5g5>WJh=hvh`v!aymykCZqA`X zmlYxm0~fZ8xd=?&@ak2A_ghORF5J`s-v0FFgtyr8Dj7=6(}S-9Poi2N4T6=2nX^-6 zyCNNvzbxf!J!13_YDnMK;4)w)lp8Jci?1&~dOMvO!E8rYxQ}85;BNqebhEO(`3Z1e zg|?2GjhEF$eWbEAiF{FNRzFA_+i|HNQ>Wr>07e`Z_@$~Bs7jFtq2vfC$%eXmRu+?_k& zm(S0e#J(Da9fx36?$2*iMO1V_G3ff6YgTLq=NW#Nmmkb~=2Nh2(GP5eF;x&UdN2zgB69Rm61GK zqwwAOx8oAjkN(nlN!LFoTmGeU^4XZQ>doY7ycj2_FdQRUDBouLqbFbm!? z`-Qg&KR#>ByUUR4%=Cf8BgBqahTe+u`>V8ZJlB7AH^3B(!i_B$vonvsA>J3q^-sqS zC$bVRLhwpZNzrr=UU8VmCQA%8@H>dvtQ76|ncSwpItf;QHNeJn@lD6|e7xMWTrfU2 zlObIsSI}b`(nxWeAk~8SQ}vA(2z2pw^*NMwd$=AL6^%BDBc^0-&cy{Qzk8Rco%wKy zeN^!Ed%v?IJ=3_v`wyk?Cp7a1c1rxLVkmMgG@7MU{ z^_%6p8?NjqrxcB;2l_Pz7A(_wV#ovZW-The995ahJUasYtwL z{zK)D_=YNdJ7g&$;1{!`NbJ|#(9)$lS8 zT2a3FNc!ObvH1VihbXRzJ060en+} zPE@w1l;SoQ&Up^yCYRqVo>D6CTS|Z2X?k@g{Dl&Kr`~EPm;6Z}4NBmDu>Dp_STB8< zML|(k(v8nVwH)P0!~%@T6lKyFNbtKz@bSCwpaA-Gm}CRBrE9YnSmiQfi(rf38rZlb zk&it5M6{994Rh=qKv@I{(xai&A#xKoBekM(8pYgZJ~3y zs1X70&BBcY$gLXkg~8z>6S_N^iwr&7hgKslD~XB7x`)r#LXT5~!;K$1gTQ_ipc$7u z`~v#abXfu5R`VU6mUV<)zje`XEt?-5nc?+xor{zhZg2_t`8^Vny_&4WX**G(mqt0+ zQb{N2aHa(LGUN10m1Jbd(iO67si6 zHcp2b>kTv1mp(2i3t{_4b}~aNF5bW~(n$CDp>y*Zq1kT?g&GLBku}?XYCPt%>vutU;Q(?mO!KcGKp zx?;WG|LXP{Q!QP1>V6~*_2OUk1xjoEBb${>G%#Sz-)xQyQw zxs(;8xVoPE0K+Vi$Y6rUJe;}Itb!0dm=c3h^iUO0adOU8a}3!18Uozg7B2!&GKvDo z7ZR8w>K7&l@0zh~V^LSiM6T!}yDx8=@_z(>j83T-JI9O>xw;i6P1h!$-gsZ_GUb-u z4u~3;v|B>N{g=kiKlGxMFEo>eOH>Wo9q8rHl6Ej+p^yVu;XO*4%+ojugz(mg-l+0y zEv{O0)kJR0PEyxaz#DMH28+31n965Tn}SEd6LIY3 z8tt|@&xH|dfCy#GOihMaak%pF9p+18Ss^K~T{H<4C z5Sbb5ICt@HebJ2eT^4dP3Iak!iRN=tGJ)wR0KPuLh758ohh!?{5}3aMR(jD22h=c0CUZ zt|6MO)9XL3L1bmbpVVN!>9W2!bFASomJgD0cIch!3t*91Y*F1UdaA@*Y7oouCUPi}6|y z(3uylcYN0vFnD%&)dm_Z!gACSESCOK)GdiaAC$XyWcRmABtiBC#neza^!twKhTlu^ zZId)7M~9^U5~Bw-ixP-BOM*R`lfwt@1X!sFFd6p!MIXJUk{FqAOuvjD<)?zhj2G(m)Bio?zZE#pgnUvFC*ifl;K9@0cKt0D5i zD&$_`gkc8Qc=vhFZf6&yD8~Zysvx$MfHajm75CyPr2_)Kn&9~DbP)%86M}uoOu^22 zGs+CM-6?X>*ojcWhMP^#Xj;jxVyv#&`rIzyQS2Z7VcC!77a3FM4cMW5Us3H zdF5(6NxyD2h9GL)gp1l4{2|k&v)_v`nEQ`SnGrA{=CL|_v~!MCzX|j36g=~jIW_V} z|BwKIx8m0ZT)zQgaCDnK%yDoO#_&*CiM|YjRwQtFX1;=Dak_{jE%h(O+P-HNg#xZT zwlc)&KvkVj&D{DJ*}%`JoUrI~dYBQ57B8;#j)9Dv^(>O^&&p?2w`RiQN4|YSInr}O zE-;~q>!ycJ@CPjeql6^$+gPR#+h(nKbwki`)sBL+h1d6(g~{P$7Z=>p8u9e`Z#Z@g zSOyq0ldBhtAD#`^NDMwBA1}T$ul0iQsBBZ;Sc-$e=ZW4t@(3L)kiitbwXMtL1}b^=J{2~})76VsTH5r8MCXS?Z;kCq<)f_p z%|{#zOHGzR6G|PE4x+Qbp$ny@%D+*ZD(TGc435ywL<(B=_Kwx3*99q=ITk$IhO%d; z#p09x{(4pgYL!qIc;YOMP`Pmh(TRgtzak;Fd;q9039@tq#SPWN;o?>zFgX-Tt=~Lw zbdkILY?LqQx>9UP^r<|xLSVPj*w1@fNRaNm~!`czPy@_nF1Y8~P+9gDXeBSPr2$k!ay`oGG zl@{Be*xl<=TIqZ9V>D2iGD_N)71Y>!GZFo@z9g-J;X6KgL%5MpFRc#7&x}N6BN=Ja zQ?SQ{rRmLwYJPkjKz^twe_>*t@H~Nuy1zI2Im1G7*v}Bh4wJq}JK5MuiSNVTLf9#@ zbAKowu&;lhXQlid&y3wbxWk8~ zvA1f&iA}-+T;c8hI>^uWR$4eUl#WnL5r*4_U(NH;mP)Z(eM*<w zUGhGr$Z0Y5yvXW$!ok^iqpzr?v+?9IY4^7InNij4wb5}Hw>Ib8T*jTrzG7?Rk}xVW zNr~ywc_?z_?)Dr?yocO{v+NX}O&p;njpxM8*~eyA#-rQGO$+>A%CRUr6Lf|>8Q#DK z(D886?p~q$)?UPC-vQAQ7BFW1ru=hk%l>KFX3;#TXY51+uIO~~q}>4yEGq2uj?4&G zTtnz|Q#_gJ`h;9JcA{9TQpD+6tEwmHy;0~#+&He%P4o)Nr%Nq?!ccx6W^XF6Y!vt7 zqRsco<{*Lwaf14k#=$G8O3kI|{MdH-TH%A1{)!mHemhJ5_{nd4uqlTZYnp+Aw4l70 z2FODg%)>$TVRg>R|GL;$0Gxw3d#yjaf88cY#QDGBYNn}Nn~JZkcTP8&3c1#GI0OosVDY&_ zLVEIIH;l11byQkwiTN{pyCS~Q-eDq|{vWc%0=8DgP|$G>?rm^{qa*%zneG>X#d zy$(ix6j|YtkJ-x#Q%L2*(Em_>+tF{JGdzdqohYKJ^hxDW3;?f1Qs?$^5KqNIw(AC+IQ3EmSatjU%XrQk?c!=|7U+lwI;G$>AjtL^ zc36Xg;y;~e%XSg~oWC*ymY2t$!1(ogRG*9GfqD+4=I_np8DD*J>c^v9!fG7KP(mBy z8FpP8bLZ3U;Y}y9^k=b)sWuEvmMPlue->=-UhFBKvFAKHQZLjGrFPxFt-&;XGrVV2 z9PP1u;MqIX?u3P+2Xx4M59>Y+KXJ0RAB7Q7b}4oI#hiXz`}^d;Gq}xL(atsTMHZ@E z%4HXIzJZgIMO?aCyu}#&R8lw4R4v{T@mlBEUii(r?8*>O*Xz~%8#460yq_vJoai4CeEb0Nm}53*1V9{%uvotCsX)Bvk6$q>r0p=R2*46t5M)Z zaO9hQQiU!Nffara)!e_7ABFkL0uG?$sheFOt<=44oodG-&wXxtgi|I?En|j8fy?4s zXhvV7ssar2VVLeDpz=wiGSVsmyA@}NGD8ZxrRxjNFV@C!)ZvXz(wT0E39LEIOQ;yr zD!tAzM|-xClJ~QP0O8#?4cz*K*}h!JjnUx6)%5hSPM}xKRErbuR1QcbRMGh`JIvj_ z#&gzG|M+6>vkC^^G3N@2&}5p9gKLo$<`1>%9~YhqKc-(}dQ&~qB6xgu0FwUhdAau! zI`;yEAIpN4b!#t@;{~Aao9w`+#@T)iD=R@taDmy<6ndwBXlF`@(E#;Kr>v|bqKW?> z=9W!|)0#6Ro3Z^r&;{^d8>w8~%b)p>28yQn(wEHVz!GYv>24W!jGsSJcUebk)W$W< z-I^z4k}5z1Mp8Td&t;}_BXS*i(kkSHl%`6%%U_=uUNxF&4i&9sYfn|F?!NGH6Cnef z`VxSm7VA?V@S$(Y8V2-|lP;@DQ|HjHe~F+`VSQje5ckRXwBL_t&uyw`U{cG|mpg(v zg8y)HB1J=gTQTH9F1*Yc8;CUcM!MbtY?r1qZc0bmch`HAP@>K@wF7&9tfz(-cKKShE6=tequgka-U!2X6EjC<$;;fv}w8DMn<6_T!Tv_1WzX zmQQ^6Qmw&Bz-KWzVI{56jBuaXQ{7!!z6m(?bDdLqERn^X=dGKcS!bP-Z~f4}aTM!g zQgJXHrp?s`!qoPC->XOK1@?en!l+FTU5L&ndI5@qIAYjQ-eyMo6s1qhfMWI0$*}6> zHzV`}HhnO-&o7s`D=lBMheo2gvcpO2{25dq%oD|1vSampx8@T}HC+^F-|kNV^s^*N zKrTX6Wn^WJ{@><^&6Z+5MaIgS#B59qSbwW+!H6GO=%y-Dn10E~Y?XYjW-{*Kfo$8r zPm~3ICwz}y=t?h5nf;z_*8o?x*TkalelpKK&dE*5O^Ah@d@;yg@VUMN6*xpG8bvXC&@_h0|)puDN0h~K;%Vt(E%_^ zdHct}#V8=gz|ca-DT%fi&6**o+&6tA;g&xK3MgQV_;IG>%j&eDG=4P-@)w5IGDh_6 z^j;Hs3D~ZQuFhYgBl~E!4?xv&$XHH4rR%;1Mwy5SLTprFYQ0w6h+-%iGsW24YCQER zZU>evF&mp}dz`%a7x{pZDuUa!LKJ&uIBi`NK9lZ@zbEM~wMwQ_FX$Be6PobjFMm=M ze98(7ybB`+5@+X7-5-f37QB%x9hA5D6?yF_H_ezhhAjX45b^hhcG|b0ijnF$&)L5y zO_@AWRLZE5Aj>fcnmSRK*f%^pQeuRj?K8}5eNsFFw3T6sHYZENpR#C*DRiuNux>Oh z`tP5w4_G!uc&t2Ip-5QW9#FnLJ8u%+ycv=jA!UJ!Ol7`X?|lZBU(}-n%IlL~z8=O2 z{OGiGcUjqoV>-{dJ7++cu(J?BF{13H0V?ouVsE1#ig|X7_)PPg{3eBT_Wn|6wD*H3 zKwHwzHgKItoaKvL-;W-Wg|26Vbb_)$d;FCRlo;gy4K>Pn<)(G85vOpIVDhA~`y#GC zE4xyBjWy_GH#BdlgYH^RF{1dV{OZ*Adm+hPwGK#3!VN=w{EY0S_QBtIz`{kCBU`cp z4B-m~R$JEHTC76yx*`hnj#dh|ujUcuBXAh`QB5hu1k`XvS+L^X@I&*O6k*RdPMR98 zzk^Aqd$s*FL2a3$T)^N2n7iCM(zTyjKH{47%$wm@1RC4B8(3zvm@`fv$GG|(Y*J|5 zL6cq?d$=d=HMS&VrO}1-zdi3{=H(U7oXobub>pY)_CgHlXVA4(>B`=-oKAgMa? zx2B5NTW)@>%yfEvU{j7T9uF`22(fg-MvWjgK;y>*}F$;O995v$R-jhiv;%+PhIXg&nZ z4lI@i*wJ5A-|=5NkibX&b`Y%MU=7Y*dUMfToB!gL`W$Jl1l)b^H&rgsRG#vHFw!AM z9!$Ku>AZq!n3MB6S(}ezN4INxTw}Vx!T-g((%c&}ow4Fw%6;1xpSZ2%_iWJ^o#WEs#{ z^GQS5lAbUtgp4{sz^oKYT5lLf-#fOoPCEI>;!E~LVl)LfIWODWiLML_vaY_jc6nrz z%Eg6CZBmv_Jr_Dp0)_LwZy15C!Mf?g?<28%MCxiNd+M8U!sB++y!woA(~{FHvo{Xv z);(U&mKxwT$!eCyxS>yxZShJP2eoIhHB020TiC=b zuXRA;qxy!L%_k|9kMX5E){=-!pPTul1E(*U+Oq-Ya(l zDiBCI#t44Yje%E(rSSNu>*pJg-uUjq8|LoP0%A9XQ8;kjoO#G8X5Q9z(QWkOoQ@7W zShf5bBFU}ukK+!p*mg>En8dg-^KoCW_CIufT-7((NNiFlnIwo2p}4mcFbvG8l=L7J zd6gX@t!%r&yet3gs!q*S9QR~eT_fBZH6HbK5hEaJkblR@Sc9@vUHFG(E&kliy%yzz z-cFB2XoRe#Yh)VSS!&0W-y=Dq|JM^p^9Wf^8_U$5Aj{)#eKP$dtp&T>7Hi?ugcptU z%$k-~$|ftloHePfC5L>lj&~#1dl$`PaS>xQ1i&a=vT+BFrqmqTCNjT@lKxE%T)FKU zhTOP-Ic7YW95#!a#`V6K2nI9Rr~*~d3MLSlbgW~pN2%bk*3x{l16pqqie@EnhR0}9vqWs+F6(rOgPjn1C@ zii8#F!lmy($F=_lA5O%c0?5xo+dHlnfVfCo02phLqj(|K@q%s5>fOxb3sC zh8?o8x$v#e=JGy>sJrj}^ZR3DR*9a_?N+6huiFd$<_47Y+}sDoNX1XesX>`RQU{Vq zbaK}MjJ-5p=;h1=b);V_$)O{-$Z({aRvVgoiFGY|?xK^1(JwYo-ikFG@_Lusjy{wpOTc)+v38kIN7pmYD#g%WlkQ6%=+lJ# ztZvO925B_=JfB=tq)U_3&M!|&`H%XUl*H&YC_lMxXhs)?(A^7Q`m^PZ}R9>VfVH0ps zZOi>Q|+a{FFtvo;692u&}=ryG&=MUchxzAovRBn!+HA?MIw9=)^8@A6glmUQYgkRwMBw``|=$e-s7|~wU*D(k`;Sr zYo-h@`OkJxwn7i_Ui3w0aXs680e^vOk#jfDx);^~kM{><2A~d5XdvPEFltR!#0r%| zrZu@I*7SaN(y7H7jguH1_hn;JhFh~@sPwxeZW65@(}nNT>mQJ>%&KQgi`6j?*9A<@ z8)h?qQ+pnoUJNw(L(DY z2}U%!Ru|=>aIygR9fduN_W6HfT*zQa;mR*tDe{i&DB`)EjlGbNx?XLjfZ-YL8~<3Y z>iv^{nJFqb6C2_PMl0+&lYEl`%mmezPA?-$qj6XMWSrlA`sY)eq1x!4Qj9DgI$~X& zm}K+*d<2E`{iiy&O~c?&QG-k|3G4YOif<=H^VO$S*gI2B_Uj>`2z>&u-fg*J%uV0s z?x1n*f(U>dJ2_6sS9` zy(OE!Y)a1o^VNa2j94pE@VB9eG(@Vj?F0+R+wqVA)-@*Ng5Y$Zz!C?sth%gF(`gu) zAd5@aWKcE;9Gse6RQ*_nQBs3(Cvxsx_wA>2c-TJp+!xT#1Bf0GxSUHhxrQwSA zw2bh}?=8TVW~F>rK9^(CSvFRewEP6*^iFy6-FC2Ey-){W>5 zWK3c(X{|o0wO`Tm$Ah86-ln8#mkkvkMJPftfE{Ua)5yD?CQUrv*8|jr2P&2elQ(_| z@IK>O>UT!xALA+HzIqyZR%3sYBbXdxr?a7vOQ!6So^t`47X(KCo%^MZx^7}fX3~#m z*dJL5W~&H}1-gTb(yq7K~r&06~&XUS{tf5SRM=7B@WQ<-%y^tcHM?TpTNt9GA9me)(`ek9V2`71}6@FP^C_6T+0gP;>fNi); z9CFZRrub@DH88u(pTgO4Pz)hGwAEA&@kiK0oiPgDom1>HX2q-$U~wZB1rA3%tSGIPq2vqa$-{w!`g8J__RlcVMNqfv=2M0eEnA4bI>S5hLlG zydtOF?^M5$hAWIi#aR&PI`JigWhCEK7=9}tGoY1orp5d~1@!iVi1`#TaC@=a%QGW| zmzJ8YD&3ss>+U`elXR$Lj?Lh`NTI}lfh{(?jHEPdgjT}vO2?^kKrdpP4#Xd8N3qe+`cb*M(sns`T&(w8oZ%ngBFMf| zYpm@;dkNAmlW&u!2m82t0_|ev-HJo*ibr~tNB`Ykb`%Qcf7E2{WiPV$%iB>eQP0JiNDyyV?Wq8UC48msj-gl-iw};8B5l<1sKj1s3^WXpnx^O zDC^J4c~IMV04!$z{|8&Rj@q_3I?Pp1k;^@431ya+i%{{6t9!T?v@cMx6e@RWm)iX5 zXG_v-H()CjcEs)(9uSE|Nx$$s$8~N#!?Jfi5b=FM%%6R3?;_EMhO8b+Xmb{j!QV!U zDNYqc8w8Lw@|I-ab(t{ZuEF0QNvG5V3(YUj|*lJI@!N@HQ1r}J-y*4*jfiutQxh!NsA*!@T zKUg!K4Q?Mcdy^(wdo9+-iOL<``rXEd#=EL(BRU9Us%MKFVgOX|xRoow-~MEKfvYX@8+BW7pogs^?}i&UV+db2Fz?Jjm_XcSa}!xm#3t_ z9`Q##`|*Wh)@K}M&?}2;V?jR(52k5Z@J7Nvg~H(LY1lnCKOE{6Ru4u~iShEx8$IdE z-AGg-T5a~ElbglX-4<9HPNe(M)*g>Y+M6ZA-V24|jAPxRG&J^7^6W{f*BJ+yh~Ykq z=b>^o2u!=L8s9dCM>D_HEqH-+wlJ?Tt*ThRDcc}$7U9E(y0xjd79m`32s)`Y#bGqP z$cDBcq^4=`&vUuKbEPOQyts@tjtwF<6*6Z}g@-1L?Ha?~(ZDl5O!LCu-)Sa{tnmX! z$#a;@0{X9Y0jLhHk{-{pQ}L$cLsLrwf>q|+hgLz-#kNMM4>kd!6$(iJ0Od4MCCnNb zdI{Dr-xaGv2HG}OK9;`ujv3z&yLF?5qC)GZsG$6Qu^uzP??a>y8!AyE8!!dA1)_qzebZg1R26o88d)-TRTS0Ynyp_#of7>T^kha2v3HD%D zbK-X*0wCWV9h2vde>11X-*kz)~{CHn!0`aAa0=|eV^)gwLC09 zxxxw8-i%1{@QdL5|IqanP*HB(-!KdIx}u($nRcK(7-8YMpr{>ORngB9ga%k>g z-IbTd<1HbETAO7kAw{RA8tb++J#0^E263>eGuhl|{$0LU_kEz5+l^-kR-OqaL7bK% zd+t@WrfVvItBU$J4Z;<9-F)Azq_skM7Q#7w&7>KNxXAGop}H4mD)vZRmRh4Gv=ePL zC^Dmm=&hjC!3Qmoc2n!(>MC_HjPf!poPtLA&ULbpCA(anC?_x8dMdvsesWboDm2fy z?=i)BIn-(yZx$3An*(Sv5oBOcn187&DH%12k8kCmii#D+2i4lC2kYhhUCUw3ML>%z zgf+OuhzrlMbNilyjUI)L{mW@YBVMgM0uwkp=QgPE-~CT+j&CMdqVBZ{?+}tf2Q&~P zsPnrpmMHM=i!l;c@1JJ;0D5D8=g@Z%u}ObSe(%1dhV9~aIs5)R59>*C`GzXnw?d<` z^e`)ivhBjsvLizI);UuWVK=GsAZ*_zTOb;6LccRg$4VgjY1$B|*7g^}sQR$l8bx$` z(io$wR+LVGk~s&Y)nmBuN@iR^1emMt~of>xtY1z=;gJEjVUxf-4&yIc_ukfNH<=5^U zRw#kC=y%!_XEJ(~kC!K0 z%@dO3#q)I2=TLKDbCP0A5=E#Xu_LTh6Z>8m8GVf!<3Piun?~b-8iRmC=Gv*- z7DKuF@?rD}|F8j_*oyc)okr}sTK1joc@%MKZy7=4#_2}o6sHO^&%Qes1*ime1M9V0fKVfnRTq(Dw&l>IK>q;qI z6z3M#^i8)#PnT%9ux(I!JvDDRP9q$j`SOS!XnJW*zca5ev!VT?)ZAK!7q8|W1URha z3<{aQN>+bM=gOtP18d2LA*B2P`~s|TVczj}Hi&#g9du}TFStGd^M=)8?q|XGLccGV ze`NXS_-(50p~=1MP`S@}xwPfh>I~mvM~d#nrt|S4L7{zs}vq)-CKzgGxCqT1D*CuRL87J0i!6* zq7%5NFnaTjiwYV?r(o6aAEvPqJD4L_nGB|zY^UMxV*>hboP`74&nJJJh41_@!3FvE zY0M<_zrxBI88p4i>~@9`E2~R6Zg!{_F8ROV2w;pO0uq0)vnA!5e4u-#ag?VN&|Wc^ z`_7Avt;)tSBu9gGFgqwe+|UQ-Sv3IGpMbS8y}57|6XLl9WM<=Vh#TY(!H9e{g1x<3 zpLapsc%?5b6O#`(vMPM(gwEfFy1$R#UDui3w%?*f)&u3s$!ASVw{O{yQJcmmKuGS# zV#T>UgN1UjX$iCV=cGmy7$%s+#GN(sMx0W|ZFUmr)T3CP=kiB((?|T!9TNR?=t7Rd zG4?`i=^Ndhl&8-_?qTVb&_6K6`Q`eKe*>o-#xz*$=k-dKmLA>ev0BNXf*VAnpCcL1 zrE>f!4W>|BfI)$Xg^%-6y@`evU`nAWzy1RmN8T^n%(@7mMqcRb*ORU>OB(8Y9ou0<#U{~i| zcuF=opM=TfSx6!E0-mp|I);HG>Z3=UA|Cjs=(w~hS7LkSA}X(6vSsu$x*ywbTrLoq zIZTBs7&LfDy$A_=s`vzC5*J6H_x@G>kN1hf;e~3bP6_N3>jN5}6+~pt8hZPv{VDZ= zq5bkqfIkumt@s?FZt{ZxUjQ`fDbeS9)Lwug_eH{1vHN+IS*6zuM`ADQTY}>R zMdR8D4zXxVz8u^pGu3BhwGvCJ`_FB(Qfo_*OGH=?h=iiMIp47MAa7kT6 zfJ2Q(#P#bph^F&N0V4M=g-|f}J?=Sq3gptCxznrw`lxMO`fMd{{4qXQ^jMQOSfCn4 zvUd$!^!VQn8prC)G>am7sY;o>rv?|P!cx9*s@T0el@QPILhbL;pVR@CXS%5gJ|^lM zt-(p@N|b~wm~$9yjsDax2SL||Wt+@0#2X&cHOPO^8>UBm(`_E>=PWRX3~WtL&qpPx zd1b&Yl&p?EBuf=665271HfmJmE_2#U3V(h(!lHtz2!Kn6Yrl0}---{(tTvltzxqSo zK76eTG$&O~^dKnwqFHEBWUki}X5IUe5%Mmh6azc2CX(`Pl`M7ngO8)7G^yc9?58)E z0$af)z^&yOymfodq-x9f!ELi1<$6!|X);9PJ}6AB>vgkp(Bs@^oMl%Vzl#wSBzPCB z^By~359187LSSJ&En`&$jDjzNJ!FACoQ^cejXaUZ?~ot67o0E6C59yB%AX3KK8Ch- zh|K|>-nIkEoXO^KKLn)AxUW*h5-2teLf5`KMM!@;GN&4neM)_lWyRtlWYI|l2cNnGsX zC#>Bs7QWf%(!Cqe`P>O2*!LhU_57H;vNKUn{a!ODlB)TL*#3p5Boqs4xU)xA(vTX< z`6V*cB|S4ef|6Cb&a#goD!D+R)}ERmn)zJq852|Y*W5NH1C7sB?xn6pRy!EWug9vN zEq_9PW@&7;Rs`g3F~+~jZIaOxiW(!4H=krjB`E8qSj-96yXN^u93%lpFEqp)+C&u=`EfE2{ZD1tK%@Q644zq0_-o{1j!G$|qEds3sc zHD7=9Hek}=?+GXd0y9QO-Ffs?6${7r7^?xO&6E>BESV7C*!vg zQ=;8OQuVK^?HHkU!hNv!xKKPfBO*J;;Vk4dl3thxJ@1@*OwDdLCVaIaUo|1?=g&WcD^NHjQ@|&AM634 zP&BI`H<}*$#A8A$oxwGrN6F3MWF<>TYLQk5*hI z@mFX^zf`?`UT4=&5>}={?3ZrygJw`L$UH^`xh`@(l(75 z;y*|J%^^^U9&#<*@(TP;Vy1BsLH^BoqMGE_`wSvFw$_Iu=3VP-MlioM8eOJUsKlb= z(8E4zlr^oIo3=-S_{MIDJMZAWu@a}DwE@{?&?Otp`cY#mqaa$F?1W$KFnmM(iG)fvHh5Dv~g@2izeC(($B+zrbA~?X! zwdqC?ALHNUfWMGc5$qV(``+uvVs35y>3*42;jVvYs=-l5|K8U)2k=j;=dHk1Tt^}p z=F5OIF&^28?7WbU;MD&kmPs29wkV3pH|$73AE4PxB~AMa>f!5x>VbK?*8>GCoj+x^ zjy7eESiap&X!hiFmU@ujzGG~K!m=$(26wPkz%pUidYyso0C{w3_IdaNEE%{Q^qnv+ z-pnAZr_yM+q1AD-*&w#mDl<_zJifSog>EC0=L?Td;IkZ6_cA2~If#m|T$T@^kh}uD zb%Bu4%enqvUUApeMmV4PBVDFLP#hfYBJ&lHdiWnPAm`)XJ?C}2P8za`or7tYf=lwV zda#&ZXNU$%&+vge*&pEXESb!}>@AGS^inr|2ws?_ zcY`Iaa-;_Di^)?=N1+b?c$;A4ZkxY+Std|V4-%T+^T?BIT!i-J53w3K#M>JuYYzHc z@kU(#dg04}XukEYT_1VsUM4pn2&dNX%tL2XZ$33vkpL!x>eXJQKYYoyVB%)4?zY!lczb!&0U-@g6eGXxC* zoD2YNc$1QdN~^gnyx!-Zj#~Ms5Jhcu`1D?Xo6vKUa2WWBBJHjrE%UT+NZf0cdP1Y^ zT9^vtndf2dSDCK~4vEm^8h7F>XtC9elde0rK__qj0q=5TbUz8?p832sxJ8xVS5KZi z_77R@S;UKPECdu!MC;P!7OL4{ddmDAZ2;;B3YeUBDOHYzt)SKZK%_WXEP<9=(Jq=A z7EX>tQFED+CW5`?5lY}Ukb=gJyj1!j!bk@r)Bhvo0O%bLzr4racI!P6_Qe6K?l*k- zJ0K;LCNXt&NsX+rRbtRN**spkusIqi8pkyIi}IT`S7Kj%r7Uf{|H%WuDN~J3dFw-< zNV2eS7gC}>+uijP#&KiFV2?wM>a0e^jQVY5GcFsRo|B}6TARIn8$7{%@ z%VUgK)$%WU>_!*EeueDg<69;(<0MDh1m3vHqY=fs>!$26AClLrWt+5@M0a##>9j|5 z{0*Do=_GV7ZrF(VOdbIT8Z_(he7Ul~y_({Mm5X-&-wYhM*tQKgBFNq}=sqLEH{S8z z3KLNe)@aImRa?YwJ%knz?8r5mpEw+yF{>M8@X2NMW}R2GVY|?kZL6#42oH??jmbX zk{9f;7Ds7U+z-75ZlmwCOH@KgDtV>N-lAfqa#3EDIq^D*J8?K`AMk#n07eVH&XsWI z%ywPsiD&ziaO&){m<}N1%gqSs#~sD}tR}UP3)Q5KPP;qEo@(S&KCyE$w)`geXoRUP zUl^FK_`IA%0K{fBR;QgPifAj2ZMH`HRRFs`GS2iWZyvB7tyhmS!UpZcdLrbQ*yK*s zelUeI%ct+UVi3IW(kg_10XvFdcd=Uwm{LFNkJQ8oZEdoviwAp4DFpV)AtJzY42^LP zq$1sOBTa88EQwN*TlH|Qn0h$(uEWLq(WUr#_3K*ZPs0NJ2r(FuFFAU9h>gw-eA3fS zyhfRjmv2U!doI5w_)AaTfMp3C%n|q(YS;S@>Q1H{bmnJ`Z>^F&$T~_qvbcQ#7a6CV zmDNhnohx>{njQ!6DhM$YlTjuZ1i$$dKi4$gmoswM84G4e>s}5EsP#kl(=}+A&!7|J z$eRES$g!Tfwicsb9lolL!XnU!mY|0j6aG?E<{efbr9Sn(k{-gYL1Od6%8yt_o$%BZ zl3mb|1}pm$Bv+Sw&}_Qep4@xHFsk{=?vfc|QF@&k<+YuH7|FOH;S3t+N(zfYR*v=C z3;Sfr6_WwyRBwi`I1BIu(D8^BH}`*4R9(u&mU~zG=GYTWL<4YffmyPCXcPmosaJ!UM8V!ze&J) zx^)inIIQg>DrKe1hlqgAjN42%Qmo>QLMt!DVUO!sMa@Fvbuux{ z-auCtl(r{rU^XfTM~))zz#1XHHXau4%q%b=HoWh=mwYgP?}Gnpr~lxwyBH_l!W(fExwbO-55S1~f1eU< z8~CW{n0n}Vm1vfQPTU9HuOi$3mMCVR2)YpX9Lz972uk%_t-!nrOy6XDbxLb`4) z^43^6=r9VfFzhQ*R@c@!r~KBh#DjW}6nu|s%h-%TNX>ux@=D6UItt`kcyeTs23Alp z?gM2VkHh`bC)w2qBNV}dY@t@%_dXOsV$IH@e^B=?issU#?lZ(ZC)u^h97_oB+H;>M z_Zyv}aeMen+YR)z6*c4J)7hDzH;Q0}$stX|a&G}5Wc!D#>TDX;%x^q&^;)Iy4N~!o zdOHExNNrvr--F=YW~(DK<9rO}oWX> z5KCxqc?eeF)QvwO-vW$H8S{(^?ZUVoBqjVY-Ur54*xia*=2}$;?<|^oDctuNL_RG9 zwUn~?Hu_RM`WYc10;P#azr+IKwd({`pCLahWeUjF4S|+Gj-YucOesa|xSXa}La zG=CTy+``n=Zgf*if_zCCPGuG0wdIR=8&TvZ=&fyq{d>U*kw|r_lB+f83`HBf}yGF9+=KX8g@BU&Q z@QaHm!W2>A%3`ecc;GK!PX5gS#wYJfU!7HNI>g)wYADS<_T$cM(UM*Qx`_%Rlc){F z+d8E}yT@Br(Dmu=nOWbcOJcG5S&oM?|C}2J0nB=j*d5O^b$t8){bV-(Hb@8P_!XwC zdzpOuakYah<79yEK+p8(9vfSzyqv@tcPp*;yo`p>mXOA3Z~MG@YiNkc++41qGNp-$ zbWj!~T)#X~x>U~+h=2ZuD^;wqdbI+58pJqxmjj_^WX)PbfJaCR#O@estph*HM1=41 zCFU}GR?yFYn@9t(eGd~p+|T|%eR!LaaVv(o*LT^@aQ<0}`j@6N)68-OW%Ez8y|w~R z5&YX@J*yRwYW8{D(X=$a_Wa#d)ePM_i4N$Ab|_jKNHsKk7P*3ahZa4LwuZk7pAlOR zRo1m#V^LAD1Y)zod@e=qT8Y0-#A%*2JNJbjwmd|yJEh+*{25zn(ZBWJ9H_La#g|Lm zP=0@A)9N(Vm@sqOC8qW^is6av0C$@Hvc z_`{6+Fdrd{3J@!BpQ%yNwQM?kHs5SM`4osD{V#JNGjr*QsNPNqw&73_aG-pCeqUyD9+eyS6sguS@5+*XXpZiCCsXsCr-LYRsR zYLf;6Z<-yxl9u?A1El++v{(s}y;~=uah{;1xi5o;n;ocgATN!F41OD5QV=IKFuX`W z_WaVWmw9Z^R%N)$Tf36yBkgJ38<~5GDvvZy?KK!jf6VZnZuLQ0+T+dG9I3rzy zvFjnJ^v5w0Szx9g$o6NQ6O9E;wL|;VFd+CJ$piqn92LC4{{V2m2tC};9H+^a9v&1t z@D&dKj}+3)5vdA;Ql+g1_>A)=5ZCjiea%C)wTV|}mb^3JqyZKT1)AS#Wmm>^BRQa8 zbjDFgbAU*LP!1m?&&*TxdB4x|(hIWAY-qX*wR!ov2#SM&L8A5^Kk!`h`@j6+?`s_R z+68+dC<%1)LNL~>Z2Lgp;*V=iG0sLC=AHf$S(tw_(gjmCtwaBYv|#$&Cad^(Bi=Xn zSe1aNYYxw!x=cXrayv5vFjN-+5S>J2>}j5dn22*Vf(oMRc7+rv;Di^LShx;1IF5FX25n5O1I73wyaETNX@GsbMx=Gvh&tR zBAi`=E0DUC2edy6Sk@4KdiKPrRnES661D1%1o5x!OlP89ELLsCL9zDdY!<22q<0^G zVX@9JW$GMxbSf+ERTOX7SckH6rWr%28TQ!oI^&|~zTrW-orE!S^PPWxx}L9zNT9RT*o_lce?QNg zgygFnt1y2h+^Ygwj1EZOSw;T#Mqp{q86!~LjT111`rd1Ih5j*xl%g;SF(js|9-nt& z%%E+3n}5XOZDvmJ>)A}E;}SS&cF=cv_3>%>s5SE zdhudYX*LeXL~hYOm=l@lEm+dw7DhW+v#8`A(@U3r@W@^LTcxd$EV$pD;?;hvw=*|p zJU6J&O>VW5x%lPPme#yTvnw7@)}VM(=ds`PfbBn&FE(u7C~iY}58P0T^O(rvh=q&6 zize*vu@Zl84=75MoCz@ds6eHER?k<6`>u}|0G;?xtOekuNl5F*a;)OJ$5y6h?cceP zoEgWvN?c~(&@b~E${<_|lG?!L1Y*2*%ukAK7Lf6qp+$B#WmsmIicmJ_@Dvi(!pl#? zeCk~y!zW%SHm!mxJk3p(n~=ZvyYnqGcF`)Ee`s${usD;T*6&Vurk(lBW^YBcZc8E{ zg2s$HbBZ%tJ;hghdyh=gABoyC0-~$eLplFD&+~huFj{;eCtD|INOrrZ6D-$n``Jw) z$sdsX*PA;aHpLhF;4=H##S5U86F61M7WXRumB^>D%Lr~k%Sg2sFMKUzyfPY) zc-4Qie<0REeT|$aUm)kuf9Hck(lfISUrJzXl+cYFDvweAiChGLDP^*D<&5{mUaFQF zU`{xH{5!7yr|JoKLrn0i;w4bh(^|y2C*(~<56u%E6s&+0%&hVeC^GB;eG^s4bJ&l; zeAw@%!Y{3>Nu(9%cE&{|WIc_NzP1p|_6L=w896B$7|Y+Q7)^-+f@D&W65&l`KE+XH z4}F$(jT~I6WtHO41@lib`zL@3Ad1tQG8qYw40N>NDU8yy?$fwfGeMmqHGWo1K-x{T zImz%O&w7C7_SCR@j>p7)_y8#fB@vfL?5j{o`p%Q*5CZ`;xyEArVk# z6PM_Cd88%So$XR?u`F+`OCWGs*_;zj*m!l+uX2;ixGRW#cc&mlCJ1cYF)zXREgG1) zBzWVNDp`h=|G6O&ErE}@(GwqrB%s2d$Zzh0|8)NSf2~|!YUVfs@%b8=<|q1ryG*Wa zNc_fDuME#)J~Tf1J3u8J;u1`Hfl7HDe)*VO6ZL>xrjq>V+_#yA8>z%=d}EU%ldN_6 zw?47|UD%t4*G-@uLs%jL=d-J_5Uld%W`|96plmDTv6D#Lv@D8LL?{c5?(59#r$t`q z9?>G_-z?kO^7ZaN3?Y$>Q8a~Fl^90blE*s5KL}Jg|JSk=p3KSJ>9T5lZU#%JWcOb$3#ErvDnh*R*!Yn5wsB&6%>xg#dZ+fD=< zljzArgTo!WG_H~b*~HHzQCW{d4T?r6LUFvMr3p6rq{_36JJ48;3{}g+u;@Zj6S8TI zk&%%l$JC}@UO(JsN|6_6-aHmlkz4wiF5)hBzcDR^cUeohOAsYmjYsNIg)Z z1)IaLNr&r4f%D?`G1p;(YrzQxu-M}H_}7y^2QYwYcrGctwY%dtAs^G%c)~+n(nbiv zpbbF!&p!*gGjRodvv?4$tiIBw=`XKxbWe*h2Q+Jp-*z@$jSD3DH3*yu6iH^`gVB5# z3=ZG8=T@G}N346SV>f^0-mwn%Z8^5eIHyk&GaBo&XT!OK_8W_?&iUq+9F2U?+Un@U zz}GPsq~NQ~#H%FrT7MNn66;t{WTi+O!G$GT=*$z$3F$~+R%Yh~}0 zV9Tuu%6@H)9o(fDW^DT<@6livWVAc1*VFVeA-Bh7CN-0I45ygF+V*-0!M?WHz9H}5 zT=$#x{^CC$g1e7cqd$!)#jI+#vVyQ z$0y?vMvyfjuso$`m%!WA5vL+ZN)-ZSf}*^Vl;EXauZPQD-+uNF=@;S76}T7R-#rM+&%$QhQ*#BG$x%V#C4W zUR2gTS0A)dzL;ErVz}Ert9PwBwsG|!&z&Y$Eu@tiQ9amR(XP=7YCLGu&{U1k3v|@>j>N~YdGk7)#U;Y^% zIU!u;2U8LsobhrPhhF9~xi%dR(5i-j{Gj?1P8Xi0v^)9uQQsVXzzgBzm=<{j&osw! z4O80l`*ZEmE2HDDsbdGp)K9AMOR)1F;WK$#Yc7vv{)d%Bz7S)2{_YZiQtG_5Q zU@orgcyX*_y1qI4b(yD+lQqGs*$!H(bw(9rC)nAVp$q0=#ot{1yoY>LUvYg+493z7 zK;%F1UA`-LD`e5x2<61Jv+sJ4`1fb7I8XHTsOoVp;qu z+N~n2A;x?YA))%zOcdajY(%w#LaqOb_}-v9I>_^4nAC8So7AUD^SX_~eI*3b2WxP~ zRRC;we*b}{?LrIrQ0&qKCA5nN6vQYoqyD_vD4g{6$K=uirYEWFDCETrl4=Zo5!^2I zA9;&@1}pW$R`g!35SC^q+hoL{-ZX&Bz=mI^wvq0LOfN@%C{Q{)>WByea%4G~e6laR zZOrVRb8I&i9Jofc-DS_=z=cV(=XwpF(R59dN{tx%Uc3GlO|Tfrcabkb1h-n65Bnq? z!>J%`ax(HC-F>@Gi7ZvH-#aP{zq6OmOz4?*SRZohF1I4gu+d%+G2Z>KF4osfwSRq= zkm|k2QgDY{sdBUhm!ix1M3s?q@B+{TjW(JSxzz^Qe%hUe-Jn-YFUMFM!%P_<(2vnR z-rG$_4}<)S+>R#I$OrjLgS%(`Ey~DIh##moDw2p3gDyj@GPP>sr-yDI8mx^_XziKx zXJD3@TS}(iwqBF^;|31ZV_!B=vwKr(!|`n6N*Vj6W`X&gaW5 zSM3Ap|9{AHBSO53b7Z#RMCNqqv@b@lV>nR@gE$r;Okuld-z)3$ORFy~9nR#MUCWxs zzqQQ zM$rZU2BXN3U2~nT(T0FU5t-~q|5D&K?Zb7KqS@a=h81jO*BayeFDVQ|G{(W7qI)-MbN*qi_b=2 zeK{uYPGxiNy{#}C7KlXBc#5##dC!RKTAtq8+;ZiWKbs7u7T-~+nq^-f_lP}lqt407iwGj@(yp5=)ZRld|Toq6= za^dbgmzs505R8RdMD}7|I*t70G`Ih|%c$rQ@wyAQDv2z7R6u$@yGJ~vzug8a)&Z5Z z_0%ST?nkXettrmYN4QwGl7IW1zm9{r0I^ox2Uv(&W(6*afu8-`LfmM3;U+n}PMV)UP@!IAtxvDgShHtS zv30P-EMW6AJht!A%iJ10HPX5b^VOD~w#HFMUu(nb|7rAY*r>=*FO(W z!6w8uQE(yVJ(KHRV{RbXBw*pVq3ZgjN!1^!>VlOixI|5jK0bkj;lPc&-VE)K|9RZ) zBK}cxeXKiI#H-Q4eWN+GrRf4OGJ1LWf;Fjpx=2xjSz} z^cp5gz5h47_`^l3L>{XmtDm2*u5(xAcdwy>G6c#k0oIiUlyK;t^2>LB79E1vLV1%_41RYkLPbYFgRr^fw~^7i9Z$MzoipOvMCh8un(>O5PtvsruI{>0Iee zIT~0Cb6oh^Bz(}OecWYSh1^5mg9>u{>A$VR?_*mi1RisG_VWSPtu)d;1%+wLxt1Qf zP2~p;96$J<`se?9+Ct9FGfqk#F#UN(Ac&VtuKV26-Ep{F?f7WQ>%~7C{f~Day-47lpYN>KHwQgP`kqU^RDp)%}z(^wtE z>}$@2-eK-Kc8{x6NH zY76SWO62mFJ{$R&P=7~dHxY6av$cKK!Dj4!G!5u!OZ-n?flAZWx@MR0XKGqYYOm$HA zW>gm_X1*K}OyL|wgKc2_FtmkUWyVcqBsX7DHx?iKxTmJn1(v>%`HK4hrwhc7qTfC= zS{796OKTw5CX&^g1o5`tD$Yl{UCe(96wHJ$x%M+C(<{$S3eP>3YlOF2Y=j)9h*f93 zsZ7q)U46a1m~dDK;s5vs9Yvs!yZ)??Kej76(du;Kr{MON(bb6omy5OjxzMoOPdQzYNAc;oH%3GN zM{C#@j-s1bJWm3`c#T~IW%oVL=W}~=uy?E$bqNpL(2v8;>+EoDG5_RgYin$l1`X&a zNs|}hOBmye6>y-nt7?guTgR;vKmwu^4?WxL_Kha~ay zW3U(;qjTvGGrz~ITI0Ta+#|BKtF0hnH}~(;hM5o~@0T7-2;P`3Z6*eHjbR$f&b>LF z=c@P;5(HKE`;p z{z00S48+yne|7AAJOTLe?!3xzJaq1rzwa@h__p;0bE0(Rk__k!Mak!2>)Wu!{ zpge2JsN z@41Ij@g#$o(up7p@@m(hRN;~QuaWY${ zMRPW->V5DVAJ+79+qq!fJ1{(VUk-!QhU1aznmg-Uzv9%KQ1y{^+RXd*G}R_k>$^b$ zFQ3!OIb5j~WJG+q%n)`8{?RI!FUl%v3MI)&(2I&M5oH=UBsBL}T${Lwqg{Lc zLx%!5su-i%jmo%Eq0@`6zAP?XsL7`VT*mZg>&CBP17cK zO|)sg{tHOBgg&bsTYZ6l>$*KK)4P%#74Q6&Xf8Fn4*r(Rr_>G6Ydc95qo46^lZr2N z*uBp$SGiY1f*Hwt7PXm-Xga~GwUNR%iWUke+SmkBky`>}Nuc6VaR64rXNLmlx3<6W zlkZ=!t{1G220s_y?{OovOVc^z6pYV1i04UQwN5vmUNN4AO=r$g1vO$!x*A!}HzfLg zM=LW5lhC^30O`7GPUb*E!(MM)B6G1YoLywE%WWNoX5Q8GBcHG`-Zs3fyIXbG;&lHVbpn9gXQ98BfDQ?`oz0(Xd=!pGAiN* zwjCX2jCyKwueQ-^CUfff^7Ib;u*|!Qq}BSxrh5Hc-d6cT03-W_ac8u&ujQD@PZ7Pg zVrCSZ!&E2Xb1mm*&WY`*bo7&`ivxpPW|8yKoSQg=oD6^f%?3vP7!MX>STWE@Pxt%m zuL=XA)pMICHXeWvcdH41zY9PWhH8byvd_nje9rS6w|R+;QO2RbRoRu%jRI2>A9GhF zyTC8jZ-_$Gb_uT$5MR%{$DJ zL@wWNukex-jw!h}e+{%6ydTus!<=yz@0jS}RdB+p<51xmA8yWM^7k>=yzXd5Jv!$) z%}JQ5;CL%5wl>{PvI%dDI=pdiF=(Sd0Ql%_XfO4xT(LMJCs~ZZDR4Qr%nqzFi3Hx; z{r9d^%fU+*>~`*re8Uxt3EERaD#A_0N)+0UydZ*y`YLH$>xswBcGNzp1(j>;UqI~#r3r^7^3yRh1eL6jo>%F&60sr!tdUtxjd<3ASM+X~r@_kc0uJ~^1$8~{J}d_N z7ALz7)iHdFZvfvBOK;}OcJ9x1l>VACs&DJ0pIxd@dB3o6g{5|R9dwBIohFjlwqqsF zu1u4tQJr5>8xHAMsTKIuSp?PC6W-BRSb9WvICkwem(4>V`Jf?1I3vPJy$v z@Gp3kWrA79x+0^zC%(k25~n;p=x3zrb^S(oe(Tg@{QG8b^Hq16kDq zT}Smos*kWA%Lx$Gz5h)lL0vSB$-NK9c~k=Fl5mo9)T| z2>*}y^ns;d=Pq3~ZI)PZPx==9cWw*`7E15Wz2DVyl` zUP(CijpOYi{<9pbpQ+l`o$Fn`6!v93n_A|$# z_B9$kYG5p(CoUQvBTZR+$$a`*trCuRrN6^yPol-M^a>hh-D!V`0U+Kbc@tz&bs!7; zBcp0W$v{}z#+^Yraeb31=}J~=s3QFTwQLlOD-{!f8jXX?>dd zfW7G_Cua?9WauNt*x^^H7x+T!)~lFZN5sW33tcJb&OMlgGh zXW%TE&85Rv@neOPf1I`=aq$Vk4t4`}d=oyJrp{NbIhjCi>$H8Z_o?cI2R zTDiDouc+l^3Jw;wb@j;x4Y$NTWeP8JDZqy#jt5K$>Y>|1om-;O^26}_6LGzMjdxtR z1$N0Md8`-wiOc-O(jV@EI>t~hhJ7nU4@EWBOz{+aKrgQo$Pf2wnm^~QHe7>myy1l> zGeHiV?PEk&-cta^?_$p-h2lw+T>>zSW<{J!J52?qvaOUTeCH@~P-%Fyro1*I^@1!J zNbt?&bq2sm1a{Qy#&b1@O5Wa)V{>?P9rcKe{^AJFIv;sTv@wf^Oh55bt;@~&!*sw$G1?aFe z?g~mhCHpi?{A^yPKsRCMY=fg-PKE43BQ|}qoD8sNjR0A!nc+|WP{kkB@LE#%$Z7jT zc~ai%?9RXA4PUtNIaQZ~ZF22WiLZC_&oL&6_w7HPcE+lVzM$*)>ih*i*`yy*U1Ob> zo#4N5Z9kdXJ7iIF6^RG8nL3#6$V&`atZlhkch)@r4Y0h9m1p9H8Qdbm5ZGvx@VzFU zBm3`Akpsx;JIm&X)bXmI)`R$uT1O>0Or9_+Xy#Ols&J9g-eq2mx!(B2=Ya-ez>n!n z0rv4I$Fk!VU`ZUI$|zTGN8x!NpLfdloqGS9yIAaj7H05fj&q3F>ZD}DZW%zdd4jv< zDaK>6X0Hltp_N*bob^C>uO8p=)RE0StYx|Ne3_|V-BHTGu&hYf^c7^bhYM2QrIn$w z_!8S^tb=*S`W0%p{WwFWOlm=SQF){~rUsUMBbz!(c;mDuB=-&J#DcayT6SI{W<;zX zY;!xB>Brtt+Bog3sjY;F2f8#MC=1QNPCTzkA4>coi7&|EiiPSjxwiFvzJlfE(!bSu z9jY=VTk}El@m;b>ron>QVwGx(#TV9V(|thPx0MvK^{TS#V&|hdiyPDcNDb+hHdMWB zo^s_~%_?^0-&ER@ZSV*(@2($eyxjIh=~y&)iUUd^^B8#*%2hE#*UHnFIZzilow_m5 zEG{kVQdTu2Z7CgJOaT!;^U0t~HTQMeu`1{ao0?SK=B($CJk`ralOWxcxE;#!N-G5g#OxU`?bz^0#!F8ZG!% zvE^#dkcgyA_A0-SP0UwLQXYR)k!XHwmsZ1sE++9{A1f9F722$c*x$@NUJ8LAW!Ub;f?S6(evg$SBNwWoB zc-Gxx6%D_%>3=3gjtEyV`5R6Psjb;Wb*QbKxg|Bl^XH$ttNUr@%y{R?fUQvrDqf0njC?fL#WEvWmQ=9&rVN zcb>Dd-*KCu$)(2p)!5X(A=WO|iGkxpUim?L_3ERgC_hXyF9lhLsC16>MsH!DfPluY zhItf6wr4a#mKa>YiqART?+3)w4cIh)K9`tX68JI|j(SC&zfC7PI_huW_CkBiJ{z5N zlmCN;sm1Hk?IuXU8+Y7NdSh0RZ7m*eCn!d3eRm`LToDL#jQen4C+WHIb;Sz2NUcgq zqpp$mvyciK-il}5IL@B@K%vVurRO#C`d6xclh`k}7bx$6%+aTtN{$jiH8PJi zo7IE-rpvm4ZmuBS|H_7EnP5m^zp2}lTTKn@Go2y3h_mha9ou9YEv{@$pDwQ*Q9XlO z9(A-)v)3RDXdv1xTl}4e#IZSf3Y*)DXm#7O#&;GU-%=8Em4n@?SV>#Y`^QawDyg-PO`X{e)+t(v_auD1 zzTYe$>9u-}$S{yd)FXq+=o_c>j>!y>e%(W;kN2pYod$>RHmLJy`m$SQsqy{?44cz~bqH?Q*V4>`(lKp6hW1 zyVzGwm-_YoGuai`FC5RE$zEPo5Cw1X9+OEvl~uo5e*wh*33%0A$0WV3iuP*ZsGwEi zk^e{6SB6E^uHV0uq9_O`h)M|3%%CWsG%DTQA>G{rgP?+dfV2pR#DH`UF{DUKcg-Ln zH9-$O|LuE@o^yWZ{I6?&m{0TJ+0Pwo-D}+&>(jR=G&^evmekr!Q8Y5j%XmVNWXz5F z$<6 z=?u4KE=aUF&d)0OGNzqvt5!(m*K!)t3H2o&rdoR%E}y4ys&kld*0>}BA1fQ#oA03; z6PptKV0<3Pb30nbjOWK%UuD1lo4<$+ZUC98XbyelH!$ulX)Sm6S)G1}eDLlu3VUT{ z=D)AH`(@*UNH0|WkFJTVnnyd?TMgRDva0cvkjIVTf5TdyqBLNW;!M8NG;aKZ%Lr5d z*~GGIMql}o0OoZv7@|P+>84@3BdGxI0`WVc#x_JO6@MJ1-Fk^yV<#B{*{hhvV2KxY z1*Ydn86VPGcOs>b%2AHw^`E^9noxwgt?|#2?w!FB&H41MI~n%v4opuT0lf{2 z8o{~$<~2}-{DcA$Re0_Ge~tFl(}|7rZQ|D{zQ$r=a$rT2siAi9sJ4rBkh&qS08hfpMI?!QBX@Pi3-%w)Cnl8^7={{=eQGml!wnnx`xN@j%C2| z)u_rWFa2WQ7i8&|ZyHQoF?LLy2$hyA5KAHBlfkmh>{Knmw8G_p9pTI@9 z#fZZ9RHt++W+UGb29IvqpDuYa;8Sbj7%oqXd17c#*05KlGB{rCQcZVcMp#&-o7>ksd_3)fd|x83#0S%%4U0 zzNW`nPk&O~o6t(gByI<-9lXnLvNW=j_TP61*c{dI{FIRp6zGIT9w6sDt~m46qv%N_ zV&*4Gyq7YNqE$=a;Ohl7KzdpapZ*clXcIEn<#gLL<%~UQEIFu-o18`c8h{aQe4G8o z@*4+)>;Z)!@YdhtNx*?L|NprYN1$t%Y}dT}!}?k9;(?zyy#K(b4WCwIHR{&kl7e$o z^M4vJkhc@Xmnsx9$J|$NxlSs_k0il@>_L5NKAOIzHU9s2W>1w^G%jNel0IL&xyq?Z79FMHLdO7}*+wZrvtK`nnG!1|xPnpwNB?wYm%u6%@&I>tKmka;XS zrtX%FO@*eNM)nQuPOm^6zx3%5rECvbOI!hA2`I_hHxs#p8j4d-qL(^+3+wSgxx7^G zg-Z0`nQ_TiOE&w2R+doK-q5m`+P?}QqI5pG--eb%)}+_lGoLx;%BfX)P61K^>)?n2 zY^Cb}6Scf4&>Agcrf7cG3$Di@UqaMO__hjk+88NG{%9%w(G|)+UFep}t>p-Fox28a>%kWP-@p&*rY*~?$!me`Ewu`%okKIZj*N}T@+QKoPjH#@dAJv22H zG1{mL@-t~j9)QZ;5*lc!ogA!ER=2-WzJ*dcaOC$xg4Khmn^9W6$)DiGx*d8G(|5@p z^1ZQ9Qsl7f*_TmBd1>kZO{|ww z{*jjxRmM$nR*I4(?{kX7Gs?hC;`J@~G$R6#U^J`2kT`il19Qp)#fyDogpGR5euoFQ zZmjXNWNJZ#YS+mTE5|Pbrv7qR3Al!4CL&$D4zv!cTp$0G%?_4p$_01)`NXN>TtvkF z2fm?oqc<$_!848Hg~rExZbN`-k&1ZkFgDtfl3AKGBXnSZTIZYFmw|k1h0qN1`EJ z@hv@BNv7JLJ3%n%WcIh3e#B4$bC>@6ey08^Up`tJg;fdOpfdQS?n91!uGyFf= zqf2$`R&&(?an{=yO-y|E%vg%6pHjn0kgMc$b2pk%+Q#t|yv#Hq8*uW~3q483!2VIC zhyQ&K8J`BAT|-}?_<1w&kQ!^Hs0$*I+HS!z{Bda;3_RZ1!Yp>Nz-f+Gv6BX1yKlh&?1XV&hi;v3ek zDdFna*^5|;pLzW|ASpe0Rigk0`V zuf)@h*kV6ggj!=n?2FQ1_5^DS|JA4KLqeB);5 z>g}QdDP`Q=B;lM|)`P4rePPgQB<96*`72PlZc6geT8uUzkG;mPFo8_A3#-58X}r^& ztDnB8mT2g7vqZ*yq84Fv%T~HSt3w6nyD^X@I33s&jrl&}tO1~<&%e;pm)fYdf1#yI zVh>bBHW^%*ai;eWn`8g$;)|U`nk4pIbZ$hi0m!te zlQK~&2k#Y7(iXy-6&zidp0V^esdstpHG2{19E@{Jos<7-=?!=JNA?59&v0wsNIp-{ z7az7hzhrFX+&QYuc66-+_+ZdCPEni|8jAo2LR_SFju=DH~V| z$RqjHyCd|(a^42L^B@gsq7$`LTF|TLUe}|&MV^o>L$k@$2b~0C&?%U}PcFhElo`b|Qj17OX z5Ee6Ec9Qn(%sKw7G6gg8^H%OJM&K7$@8-(zuOc2#@8{DvKP@_kI5VP`|DQm}m>8T8 z(Pu|vF!FFb0|tLfuFqp)Qg}-*?BW9%xO*SQsJPrdOi|EMx&Zp2Np}?s7ObRbb1dmX z(aFC8UAtQ;<{RcUq8e9IZ-VrHx!2E(e85-FhM4wlWc&R02%5$4LY}hKUSzU3-ywL; z?XB1Dk2cuyr{HwwQP~+>h7O_Qh*EsK@(h=3Y*s(jtJ1WYw-C)&1bSH)snnb}zWDQI z;}DZhRfC>S*k8o!)IfBL$LxjvA$DH3QlMrOesDHE3tAFe|Kj}LvZd=Ot=_IPk5rju znMM_L$}}AeQWU*MVqmvC92RQKUTUI6=<}9AU`X#9e!pJ@Z9-{=HP$A_-#gUn^cqgX zd(Jdq-RU>?JblAWOI&xm=j)z=o_z~wR5)2JPw@nWL!p@ZVR{2exAH$Id?ypdc*iaa zT)ohjwyGrO!f4!!TNOOud^V0fw^UZU4-ZD2*MO&bkk^tmAXG&c0t{Q2`^((JQhQ&IIoZEo?d zo6gKGw?tA*co?VOHO%h$kVsx)#_^p5X>n=IXE+J3O1KKI%O>~XlP)ka8>FqT#K1|g zmj|i^&e+~701#B5KnWfUSCt2)?S0coY~fuD7XrB=3Si~ya91vLnBv{Bzp~XZbH(O z&3XRbK6GTfT?wVcNx6>$ouu-fc1iz|47_|fhBGM`6g7Suc6rW!v)**g^NuYV*Ve!j!XZqS~Db)uY79pJ=Dl< zn&CX|+_+jb*Z`jUgpj;(K4|=2b3vn4w!k4J(P6z;6J%_v%Y4PcVZ2;>-yyj$zNY18 zt0p62zBfIgVck7IKkuzvOSwiV<(2$6m3q%}Ce6v9+%Rs{7A+(Z^enBq$Ye?xF6JtX z^Xkn>^F2SF+<897(QmdPo{d8<;ty1SQKGhT$hc5V!b!cRLnfj@F!$EDm6o!uGDXa- z5aUbVD>A2}AxF5T8Zu6kx_p%YI6`|?BBv)YE+My_>A>YlQ|(wFiEB`{;Hnt!)w2=A zc(0s!tQ1v{H;R_2HcsRU2%0WL3tUyiTMW%RY**rPk6 zbHbM)`nJ}Ha=w+!}`5M_j%BkZPnghk|KW;dI*}%E?7lT zkv66T%U`a|kdWLGx5y*=P#-l6_KZI<{Si3l1p#w>QxOM}*occR-fGp!CX|a%H!&4l zR(JqCPU<^I!ahN|UiH2qy})@A95~(g%fA={*8rVJx}ehfc$+BWC>_`_RWiqU^E|GZ z#d4psFV&gr;2ENIg=$a{OJ#{dm4nXN9sIX}TRx5s2CLW4n~c1kMp6rwd%8*DzY{#& ztLT(eAdoKRVgpgGEr?`Yda5lx;}YVnWBk+Z-!)yA0Il6b7Lj~UGnhi((4r?9 zkokHrgm)UT^!pxs5R)^sZsVN>gL6XqoTRv&LMCeEwSES~zPQ-8vrIfU7kxJ|^6o-2 zRq#O(87_zuAa#!|H8#31m9=$z;9q8r)(l0F#TV-t3f6K*y-!Bm89oJ}zsSL;C6h1@ zcg98O6~o~X#h~D$ud6V*<}JAAbgo1b-?llAFA`Hw2Ah>AZ#6$j>nOYaRiy+Q))PjT zexoE1T3I)3*ID1e`SiCQ^bN!$M<%gVb8s+|o9m2oTZ94jsG zuTiyB;6H{~`e-;|D#HsLE%~pO_%kaxDM@l@YJB`veW3PF0M?(hIBS^?<_yuDvnnjl znw1L>@0wl%ng?T6tJ@_<0Ci>}CLPPy;< zqx%6-&kPkx_{e+DGCW3NLZrBaG1m9hec;wI9xMEi^R3k#0@NP$RiMu#9>uU|{Qd|S z+9s9@^=FtdxHprvIAdx##vAraUfCSjoue$-AbG&?Jh%*`t#>${u0OuT=hTOzSfRPx zSYMb@4sviF{ZyLe5|OWOIzkDoYd^NB{H&hbZQ;11~_Q)V6}VQ>Wt5lDQ{G)`)DzT5zq9Ka;KylBu-!9h0`-eizohl z1Ghg38GDRw9w!}E>FCt<&oe8ph|guD>|-jj#2sc|0e!@cUrl1<7TELM@2b@rOiY*& zd?zyL_e4GeE11OrCG!0`sp5^Vgzc8Kd{Vh#L7$PGamLLD z5L&v-2q1*f*2BcoZvg|^^*^t!$Hb0ILUhuypFDDmN-ITk^+^r2(p;!(@!p)Oc|@PP z48DJE6QZlKIb98^A=VPrtDwkriL~|2@WRGRApkSAf-DAnD8u^?y*D69-33@GHeBc@ zIzu>ROXY+>P_9SYT$$R&&a>7aY}xG>spff8YMLkMwnga10+nF~c3xF@XDjrYe15AU zAewmm=s?@kCHT#)JLpP5j5R|HG$T{BW~;eSuOjo=8A)*3>$09$c3~Ma0v;xTOSHp? z5X@mF!l2q(y{r`R1-==#l9s-%J z;gy3+&vJ!9wCeuWD<;-jtx1zYs0qYEH^?=9F8a$lUq2{yn1oX?yjI1wEH0>?C1>c0 z5;#h;kw}r<=R3Q|<~mK2K->Ng=JGMTIz?pV#a~H9KK$2=RXvyLcl-AFC8rnlnB`;g09HXCD&NTi!!1sKb)0-$@}m2sgM}V-rsd_IKzb1eY?nm#i`(Z zZLJENCZ~i-X~}IoLR3JQ;lix3Iv?|fFSo{7YxgL6*M9E;=pNXqh*c;ftj*ICT{TSz zEcQoJte>u)$xVDaZwVS!ae#*q0vsr`%tPmgJU1_nc%tq}oKu#-=F^lweC3r(2b%ql zq1f+c4hawE7h1}WYUOY1|9o{+fUqXpTk>Dx>XrFzKj*PGv?|F&$3ux$x`@LN$ zhURUZ=bHr}u@G0ndzS#c08IF|XR;@ZHU@1oN;SrO(ljf>{7%P8nSUXyeLLyJlD>hb z?{yL5RN_R$?lI2whFT6PVS<>%#ouS}?Ysw9bL%jRJ*Ztxn$RCgNIM?qH-F;M=;H6j z*%OV>?;S}R370hDI(9wSU1_S!DRkVH+tfCQ(yL>>WgS~0#G^6W$v!ik_sS8K{T$+K zYBBmHxrEBui0<2Ux!2ayrK9H?Dim0WGj99g%dV5)oS)tolj!SK=3{ zIE`Lmz<1$7W6eHJ#=amRw$Er!q?R1b0eqz%u*R)Et?~XDL7xZwb#EoNW^(k1 zSmC>k6TPd=bDwW`-Y!opoHm36&kqiyfj;*&O~zRRO@EhK9GSaJ>EGU;|5io(nmYFm z);b*OjhtmFm`$6QNvvKJdsV8TvoS5*@WXttu(4?BZ{+)<-Ow#cJGkjTT!$lf7eYUo zR)+T-Jnx|=@}(}{xpAIzm#wfxNn-3=Y%)*28V9cE{TrM|kFy3()Dyz7CpKBuGy`D^ zMzkpd=b&*nfv25ZC05))`ad zGK!{v2DB=mWr;B!k@aHwufD}a6B(5lvcK11$fBiBN3O6<1 z$a{1QYSD$Oyn9T4(!t=X9^Y5D9Qt973ej!OF-y&EP2`)ibmP}_I@mhoq9m1-TEs+L-?yE))cC&E1Wctv^m3lcl?xALBnE50-`-9< z00G-nqVuiY;YVl8J!JPJj%5P7CLP#LC|wzpL~s^&>A_f55^H?ONXkP?-#|2cq-eF% zrG4g<-mOis-}RlufZwYG)5C#0{X29`%q@c32bfxC9&fXO;=K8b4N|#Nu_tO=!|aeQ zhi9#tJh*27v)h4v;kheDAM|U|Y8fb6ae-FZYmT&f^QJd;6c#z^Yxi>S>4RaiAJ| zKAjRiPzC?Y68hd+yCq=h$Q{MJ-t3oqzRogOSi43K#p^B2jG)tYoZer#`1?$`>_XUu zjxVBL1iNYO$S%zHj0_6}B_rJE4B+KX4@LuICMwe%L+pep$hyYcE7|8==)H=je5;+3 zQd&}&D$R@SciHL7t2jD+d;Gezdv2Z2%!cv{z4eV+v=a_4@aj8>r*JW7@ab4 z)%xx`C*f&OsW*G#kE|M@HA|BGv;e3BIqDH3Fa%mSx+jnlGt=X z!O|vil*^)~FHbu_@40n-l*|1J;<#S3S|5`L)tkU@TJ>$yLwc^=wVsOEav@1laLvp7 zi|AE5JNFhnW4E=JmV5|v%#h%lqX19Qlyw323w??wXl*tuxV!Bku|FqZhCw#5Um^c} z9Non)MSI);>Pvz~AqO-6@uMH&@jZFR$m!_B3uGJoK|~q+2Rc$stanNpm_+6_Mn1FF)KyMoOe} zswNCf2_7FS7c=0~#;{oBJw9<8q6@2rpQUBEz-&FAHR!4jLYDJL_{z|g1uAO(UxI{D zB2}ncmp`gfyXvglaw-MelFAaC_O@^#Gs@&*qNXCwb<4u+S+gi5+l!Uw zRu-s=V=v*M^!tx=qn0Pei#htyY|h%e^%x3kUOZI77A02}^zgonx!6*3S&rAb#CElE ziHg7UeQ49F!$*OZ3iXr^QvRu*e(t9T8KcM;`~e&sBJY`!xb`iJ{n+axg{g<9vn+*$ zN6_nweDzM=U|KbJAFwGvuSMhBbbc_WeLr0;-0WKxFh9L`N_v+@QlEd z_qpaDlI)Tr2DBis`zP=hk86g!)EV-d$h88t*spmP1hDm;`B>oM`O>GN=3FcIQ+z|a zRAPt5=Ul~<3zr5AZrC6Be;VsmU6i3yz4<({{EzVb9@|#2^PZ=`?Mjskq{ehHBura4 zzrL$5z)KtLa8i~pBuuP@Ud%%nSQn`9GZ~0XUws;z!8-1M|dJ(8%F6sK}EN+)s@ox2+( z*>|Gr#x&~q^eXWYD-gELjkNDb{)BCOYvj0(a13n4+gCh{&ewN9B|Lm$OXLQTFDHs@ zP>Bs1iR`0PA+rIOJu?nNA}@(rp008l;&r{Cs#dsf-=nvc|a9^f$(VJ zn&eabDWUAuCc%-`>2{z?MV!|;Q-`?8jDYAtGaI`vo2I|D8HJyRFl&vghH2kkN$?&8 zf~ePq6I6qLi1ffnkCe!D&VKaSrkJ`Yu2`;Vu|J-J^{64^|8)O(z?mF@L(-VddGuSb zN%=ufS#Ff++j9i_6nJL0tt1!U@Z2aZ^;nNXKUqu^tKVTQae6qicBkInqGr$GVaBdC=*@tEpp(X`JhuifhclD?ec@Xsf(jbQ1Ba^7YEY>w-TH&1N zT3XBbL7%@hU#yFogRbN#0omoaBnzBE3?21wt!uW{7wHr5-G;90lqfi$a`G++SgE8j zLw0@obJH+C%UQ_-u@ zySdM1zqxc7l2(GunT>Q}9*3`YwmMy!hH_3Z6G!|ypbzvVf0~)>Yj3YeU1GXE?Hfwd zBaT2Q@Q+slWL4oQBL9L<@b?$>(YrSVd_xd-;D z=wYD}$0s{6wcPD|o^nFCHU?*gjssPJC#WGUKEFH@^b${c4IO6j_Pa#8zqH0fA20Ko zCS8c$VRrCOoiyRtR(%E=#1K?@R{D2gY_+Pe@Ne*@u9(_1M&IH^mzSQ>W}LX|MK?CD zbw{`Q9tgL^9Mno%y=WMgr%tBs<_A#I&gPlxg@3LgHzgd1^b~ZnvkcqPOvt3ovt1=tpO*H}j_FZg|(bO|$HXRM)vL z1A!s|otZmw&O}KE4>|B1WVe!1#JRnKs--yS549ikmLzYl1@r}8H!yCUfA0}GEkPI$ z{S>AEJ}hYTu-lZ9-44R-!o5=aJ57I-3~Kj_CZOl0+AAQwU4$)#5N=`1S}PE7J+%c3Qj`Ge zC7L$-e&j%3Cz$+T`IoWp(xvaU6L`*QvglFHtyNb01c~7j$NBr$tbT-MavkyBN7CfY z?=)A>ca8waM5q2sYKv>~VE5lk_}nvZ+||K=7S~ctjFFTs{CqE^LrwL-bV!rr_y3=srYq|RYa`w9KX5h?xK&}@?T7vhy$RpLpxHEM} z@oJO|VxMb$C&;;>u5ARO)o9ch7-KCwJ7AO<@o^GUwH7L4Eag`0VJ4Th zX_9`{QoT^lFv5UPIB|+=3aZLKy|H!%_Zo#Dw^!t@75W+jD}bb=Ubb?cbN*?SZm2IT z(!;`KAyk3uj3WW#03~)vQCvlP?+gfvm_0L`!#pDlSdjv36bML*tnAD{V;=FH+TmTf zgS-^}V>8M28 zrpI)`B@Kw_p9{LpMGp;+qRGZGj&rSpKBb^t3X2HT><#A&&qThbJ|hR4zgZgd6sK*txH;`kj@-K$V-)4p_!{O)4l4udy_(pIb;DP z2Ty(u4WojJy<|>?KF&3nRzKZv${eOnZktc~IL_zyUEnK5WlQU?Tw(&`br2MFjw8st z&%$L1!79YL$e+`?1xJ0a;0s!UCgjaFanBk`1+^Q76qm8neNpUb1gN!CVH>JHfrz|R zM@y5DxkI_x<%n4Z260)5{N+y83%RPrqQdj^i>Bf4FFg5jG^MX(eQxYlfKC<7>*ezU zB-H=-hkG1bMS{$qfZC^Ouuzi8_qlC~epW`S#RCoZk)CmZF67ZjGHzUxH(4+{L62AL z(pU5*U(IzbgGuzQSD1ASQ6>9JOe)Hy&JAES4E-M)ddbL`f8>j01KrMoL&itm6a zE;lsghURB?H41YB8t7VCy5%%kRiZd3aFB1M>LrD~Hv?W)jXyXMyLzB=F%jt7dMJR1 zZDch1BNid!U1*-dJs1)_)f-HjQwUIW3Nb^+^p_lWxjfv{rg%34@=1$?`+ViIhxGB# z84A_Xy0_T;cy7EV;s3{6em(Z3$%`ZtZkUx!tI22+h6E6G5A$zf;i-EaH{wxmFvVBm zGg#D@0lHnLQ)I=e!yVNz@4IstMZsC%5^E=|0Dj3`ClPRe^o#xY__@TS-o{D^MQ;0S z9+FDjN^K!cG8b^a;|5i-4v-ym#mfo%OR3u3Bf>r8 z_U>0l*n@fWoY`*o%pM4VikOTgKQ-kmo7D2XZ1shw%!GAwvN-7{f5KdJ#6XRE#v$Bi zdNIcTd!>NkHBT#dhQ$FC8(=<+ym#L-YwT1OM~(1I!QSB@BjQ|uo{JsUdrk93_}4Mq z<ae>~}5un&!XEXeUvg*pj6qs+V7bYRGYd0wZ0{VgMHcTdQY9O-KwzfI{ zQ908%vI&1`a*~!IW=H3~^V#GfqJ02-&`h9ojxV5?4qsYL#BBNu&L`UZ)u8Gg4O*nonB zxqUR5W?ex)GBt5#o2*tM;P(3dZm5zNeQ4{BPVx$lhY$yMu_M@m$n9V4f1NBVcLoQK z=ULvNHO18~hh}_wtTNM`rxM9$D`}a4G~Cj%>ap^1rgzfD&+O#GJsJ9pFMNxcm5Tvy zTc>DjjM^XQ4ce_Axt`Sq_a1o944iL<)=&l77*v^3f|JA%FIObd={gCINxt;LmvtUo zs{4MP5%z{z~2QGH$`2(hUMn$!&^;Y%Jx58_Xa%*&b@zZ zRwf0A2bTS-7O|e1~yL$k!XbKgPx9uFKYxpc)Pye@U`d zoqt^ZDVX{#`-00@y?d_83)J(5%@2W}Tq-IVWs4W?;Tpyrm9$GBZ*16opi)oJae5^<=gfs~n@%HgK&8N2UYiR!t)IuKyz8J&Vb0t8}h-CEe#Sh{LQws1hn5R$5t>ZbO4^zW7c{PY~trCbI z3R_x0R{I~c^R=><2rp=wIGRU5*#N^S5_Ww?cp(BaD572x#Aw!Y@U7PQT`PKro>u|}#M)fE&Zw&sK z6^lupKpeq>NY`jTR^rpbM__Rz_%Hi3j<%2AGV{+s`?Z59yNcD%5e1_`HuHOdP*~Ub z$X^sUVcJeaEhs_ZJ5EGcIc+EU7zvgID6}qi_C-#DN2*<>Z~1DqKc%+Q)Vb{}ul_Xa6GCBlbzh!!-_c%S1DQlgqeHwa`!8a6B2 z!cp~N*lDR!O*zvwmTW{BQkO=2Ufy+ElCJ-5B=>{J)u!*uajyXr=2?1k)Si_szBhYS ztWB|p^uB>e_MJpwOjY2X<+-TO-}S*Mc@QIq1@b&?&_Kaq^wK*M5Xof4NZ2NTaVzb& znf4uTd@=H|ZFn`Fjf)a2pa{g&Py#Z=hx|Y7_TF^(hd0!!&*;=LG zesws6v8b*f`4n7OcGEob@KLa_RR6eq_e*fsYbex-oT9BH7N{Ou}+>EmC%L`gqZ^7bgtC zrw1#pxaNdAOS``BM48W*?S*9W&zfWvOn}%JySh&F8_aXbhuCpao=nZZNt?;IycC=Q zp6=3I76x~>E6skU`g=c5fG|1D z5}B1=aWQag$bT?tZM)(JCKGLJL8K1eD!3noN?dPab5D&k4$E~)I#MD@{fx?_Auvh) zyFc{*k?`ZnKrKEtT30O9_{`*>Ta=L+~IS>X~X;NT7jo6$Fy@o6wpGCM&>L^PckcVC1FMaoSO()l zzril5IMa-U;5l$~31YLMCL^&|$Dm5+epwx&oxk$%*TE-5Spz-?Zl7$oqOkg#S5N*d z;Et4TWexSJ6`#1?-gb(+Ij1c$7|aVY;tsNuR+apoYy;x?*v^ZWno}W~b>*vZz3W|# z=&p4&#Bx0977cL_%HH}b*DqzVn0MHf!ohB0k*42B(}03+sxv6qH}$QfQ3}F9C}|}* z>8tphwtd*8M{Emn#xQvk$#*!t7e=uO4Z0E55e z$1CCBt*`TSR@;ot>}g#~qJQzw(o78vSbz6z31#bTDr5Ke^6HeM^Qlr~4Hgs{ zJFJU*SK~Ve$We=fx7qPWQGD$f+cgF?xrt;Df3x`3LwIqR-VK$7JUrir181B@{^ss7 zCiP<^$9&wq_6%K0K z2LQUWcK-TJzKY6L%!{@QmY_8j!=A}qyWviypkuOw7BfIK>DOX{kHcba{@Ny5K>jX+ znw)EY)AH!$)=T>qp)r|XQon|)j=wwCkZk(LXWZ)rWjV}0o4S-t@X^i9`jcV@Z@k)e zkrX%DmNF(K-9lLF(u-D|F)r22wkJ4oGqtHhGKnBaQZa5I!RR!5SQxJKA5ttyTY^97;b~~%9~)io{ou9*v-)q z_qY{KBCXaB<^_8LHO{Lx+&4Q?WwymXQLT(x&rbPpM6P3R4|u&Wt1EctnkgFh7Qr4I z3wr=@dNh=_YS#5de7m?qTFUIQ0OkBB;^)TIRwR}B`>*6Cst>8VUIJez*~1xK9*(%a z^?|&K59fS?JEqD@wFi0`EfKJaXeCD~;hZm2KM^_$;$M3>wK~pwd?4o>(s@E+!etQ% zy;I)5aQNX2S66HQ)}Y7ue6Y^y{(W@q8ttA~l>Pp|+XMD0TNBFIs18wVVngdt5%*q> zxv)Xin9t;{Xe+PeH%X7+$WZ3Tv~{O;t)yW#R8UHvTT=#k`1IBe2S`%vHL zqW2w8W$cV9Gf|X_?i-CcG?Lk#*;vi}JnL?(45o@Jw|_JI+i~1}HHf(9n%}}h*~8mE zB(gs@Q`JG@Q<*Mh3ha7omB~nvKL5wV!cGeJc%Sw7;+3ZFHXnn!h)8ZP4{}7&9|yP( z4t;N8xnHo)vV~W~eIoU4&4aq)ox0pHk4It2c<*jrVpf*l&wH*AoW8qub`pseKk`+# zALBfN)eX!nZ}5mcDzTcAKI*g8uu!4ERn_7|>12`Ovy+$Nr}&3TPw*Qz6ops2I9;KT z07g1KnaITW8~g^hdlUtCCU&)%O%huol%3v#DK>EJev^<+7Ton0>y1nhMkMV0@~7Bm z+@wqU!=0*pGtscTch*qqv?5KfM@_uysrVwVk1(0Wl*zqaWAjnnPV!}ukVVSL{CIO)w*G#Um((2V_h6a-7Elm%4~W8YNL% zd*`Lph&rq*Wg-`y$-|u>y;R(&*^`1&-J~^DD!mQv&d%(_Jb_(IVd!^WP`j=|wiiO1$3*}F)7@9~X zZ1uz{`yAr@=>3jo*BqUVMX$?zZ!}%EQ@QCq@BVJiQ&--LIjH*JIg3z8&AY!0elX!g zZZq-D5nJGj1`v8*^BW&(Q~X=-Y?JbpJkz>t$LDM~X(Ltpa~i*LJZsR}qZZJTNC$>2 zmt-+WAO$ZEKpq8?p9%0Ep9(!3wPCn0Oo`iMlE3XwL1ujMLPhOH0{PpQ|9q@&h3KHm zRrFT1I<7p3f@{|(soI_hc=YOY<=WDO!S3AZ4$n;v?dF}=R^>lDYf*>Ft#7GaT^MX8jgbY1>=gz$ch2$H$7hdJmrpejKyD z_371zkNMIza@ikh?o74V9u|5|M6FeQaVZdpLcS<~^UVik`+%oEuJtg4bm9)_oAE9! zE`sJXWGoTzrOD<|!bzxlVWluMf^m4jKOF_-lkWCFw&OSGB3Z4w($u}4wH#X9Z}GGa zMJYL5FGF>9Xx^OmGa^Hr<#jh6TReT20#-tk9t69um0hRYK5G$LlvwhuhOJf&n!^O3 zf)*#B4E=h9<>YkQ*TO*;;#+72glQ#gwF%g@pO4qr_r1s=7f8Qd%I^AtV`1N5V_e{o zs>`H!xW*7=Fll);O!eV|1wG%tybDE?s`-g4F1`Cm7r?PpY(2bbgTn1ONHP&!%3P$@ ztpC^9zm>}3G?z4W#BVi!XM5c4wCCIwxpy7A-f~)DQsC*wZ|hv}xb$axHj81qTnOr? z93HJjX;<@@YbIa(PsCAQ3z?(&&`uKW_G2_~8cdvU(|a#0OIb(2Hl@$Ay;-is&3-6b zY&*jL6RFZc;ym4@WHGb7mya(7c2qxMt`pTNxs!vPyAkR06AWtn6w9stIOQo>eh1Bu z$D5rqW(Ulc%h6O3ch5At<(OP8iooeyHgVRER_X$5h4r=H=22^s|@H)dw5&9fAcy zW%!8~5F*;8w^7Q<(~AbORjEmSu>k^w(XfoZ5`zE)?S=d7jBh?cli%6(m0gu2#NCPg zxNtM}@??#6lBEk3&OXDLR3hz4`hNOJp|e<>P=_*IY8UA)hvxc00m5|ZVffPv$iAp5)eOx}Lto6L08pyAuz};>kMi}UDf7XIJAM@mai;J(`inKakfiI4 zgI~8$V&CUuSqS$l4ZaJdJsFW`CMpk+O8xg&SxA1`5r$m(s!nO(!g1kCAiueNphYSy z3Qo=tKK53Jce+x(P4ls#rvB^1?+t#HoVsUM*vr!`(r1xGNRQm%$~8_~n`x`K2csdQ zpWIvC`%_C8c%?PxJk>zWfglh*74!Vjm$PaXxgX<^-WL9rD%CRwkJ;F6M0k6hR@y85 zMaz8kPy}kyatngJkREKul^t>EuXK3i*ZfEcYN&Fz0918=nsb7<6W%w??_`vGp#t0$ zLGyr?MO?N`ON7lBaw)Ue^?*=py01Q#>*mD)DQlFkK)r*to&?|XAj^7`IzPBPrInMu z<`KcTUh2*L#(w}}RF^!Dx?uKc#?Ivjyxmb!SJ%bf8|6S-THwt?9w>S7(D4PA$O{r^ z?M&}UnatE|rT6N`StG_1(cXUgA1IBk9~6_GOf4a(Bjh>gdJ{TVMCvMX79E)83%A|u zhYxnV_DidS7SO1<8#E-87k_y zveHYLR;=&Eh@#$F(3&9A$kVGOnY+yrj=k)9@I1k>hc+B;y;@k$(p9UEBL~zfLujJf zc?v$7ktMVLN7q+}MY*kS!!V>ur&5BH(jc88-67rG4MW2qh=hc6BT9ESj0(~vAT@Lh z-95xN?sN8$eZKSiXRg^~C+$&%M^7^BNhl#i;}CEw(*nE9w;pB*)j|AGUw> zQM@d3zVobz3W9-^ps>U3DtClG>1c&CNl6l6zAW@v5&Omdd*W5zAMBy8(%$+=H#jrV zYPa>C+nKzKyjEMvy3m@0+b?E~FT{BkN_~c4@mvMd|f_$%T^%J zRkF~;*|kqMP`mMQY#)U=F_)PGfGY@wh#kiJw@Evc!G`MRKwTNLQu&K60f_w zbLEjaABigJ5jk6yNG~Vvd|2$A*X(CD zIaulVDr)@)<7;)YwtQms2RiPV4pHkS-!i*ePj$=-dL!FnVqh=vI+GjJkjxumgi<8& zrwas)H%Bv)(yf(=tNLixPi_GhZl7u-_x;D_G87J>FS`LGltVp=F0YyQZ?N@GT$S_+ zRIZOEcKP9PdWN~><%2BN2U#OhkF_PrGk4_&_?7?r%D z7M?X%Q9X*V^S}v~wOYwcE-uyKE9qm(Q}%J)^}D)#mY2;<72k6eMLgnW*87>m zAJ|)^2XnxT)C~S~&w|RiazzN+2RQ8$g+y6trH!YVH0(`$2K0>NiY0TAX?OA&H1>w+ zVwbuiY103V-4P%@mKESObzsfvHs%p5RNtS9-j0`a9V)+P7O@QH_k-M_S;?qB#OpU) ze!FQG14BvHPSilWu95?64G_&QkqFzvtQ90`TrM0znbycCxFgM33EN9@J$^EkTcAyt zXg)xLUAN|6&Pu!W%(l!3c|J8oefDQes&$`D5>2s&bODLiF4khP8T9tsYE^=^1%>J7 zr&^o_k0q7qBG>K`({RhT4_dP)Da(yOadBIUre9|6-MA*iu#Smlj6h4`&KDd+Qg;h)CJ$t+03&HV(D~$;7EVUi%K*ODt$b%>VYJ!Q9 z*Qx#;Ai>?K0)}=MkTuQNs?JH7J^KUe`13o%qNfuP!F3>7pj~gRk8qdEu;D{OWr@n~ zFWw>DokxP4>Q9uNd|G;HeEf-N`;Tkb;$SSLv!cVO(VHax9t;zhyR+wY6c4)MmmJ{p z#Pjp`OE~cQy%UEz90M`M(VaFU(nuvQYoq<@{^-@4CENXK`x7Mtd^xIZzjUi_H4!bxp4y!oF!6~0$6C|S9nDq*cYuxg~lZEwTPfGnOzUK3m z16W0$Yu{)%{dqaZnP8ECzI5GG?&`o4eugnApR&5fHLA`|c^SY(z_-Ta$q+5vc_MF+ z@{Z}HWegpU>vW~tuAf3&T1_G1L_JRc2@V<}U!4-3mgp5qxsA83@bw?JOcM@EYe{EQP%aCiJCtV1i_rnqYaJvgRPFnBJ!&$a5c z%cXIAo8b@pWkV<&-8{HCcyd$8KG(Tvqf$$Ro3(pgOhhcPtE)~(#FSADAA24^%n6P4fSABzX9Kv0`q$7^T1Kj*r-DQ4c2Y_)Su?ms%2~pAD zE9wj!hKlTouq{g{5*TzJd^qD#5`(B9gMTAZgHnJtP_EAE}wkh zc!<~^`QvWapx!^I!D#-P!59#0VF!NsM1pgz&*GTZ;G^X8?3!=>b4E}s?lRa}LBHF2 z4dgx3ZJ8d>HS)C~-(WlH8<<*{ya6lRg=t9A61c?~I;9YykeQERtVGkhA1JS?(+_)f zPSx>Q>oD+3c(*w}(2){Nh?mt$wDo1_qmirk2U))Yexny!+K&4Z`RSSSot4-y-k$vS zC}86kbQdQLRw4Ni@0#5>#(>7b*luWV)xIr959MsG{OQkk3q53c=+Zt2GUt9ss=}NK zdFLw<&GN2NhM3I&ja!i2Q?E+?a+Op=>uc7l(g=e$d5G`Z&>D~Ow`cY$uLv2>d{o&w zUy9FFw#vTbB;3|uBEjDq$lq8VxFSa}K8c>|m!4ES&$`*GTj)@Vqs0*o-WHYc{TKdTkDamBm>|FZf6#I-%S#N)JmOFJ8RpVoE% z=8$T!@wO_)XpDq0`_+>1P7RTJk-^5OzS62%eU6}d4lKJLalVl!2=s~hF6{ceaF$JZ z8Te!PQ=JVZ*`GuW!I|~@dnf}?0?F_?|6&ZJ#0~hDRV)X9wYaMz(QaW zGVt?rz>CQiHd~6o@fjqWkwLq4X}Y24)LCIJG}~57b${;{&OU89@mhC(B8C_+_?7V> znv~NPc?)ln+-0g}kiXVYo8tN>+9sKU1mAZ8YQXCFAjz4cJuffs70jqr=WyN_uhH$D z!sp{m+U^C0e#ZL6NWt~3X9Vh7swrz0Nw+q%W6<8$5DX#C=g*Wpf0Ft#(DBw~&Y$d( z*pRe1L+H03Unxf!M~+IORc4Z0KDLA&fo~$2y%9(_sozNe4^>hIFT@ zf75!ui6A=QPWR+jf2iX0!-Ken#vAeD(#4n`#Rkwe+u@P38#ZV7+l}e%0D4p%in*F= z%8Lk*mb3fw!x=7CIzF+q-FWjwg*)LO;K=S>G17Ng6WgP(EsWp(JtYgt<K2&e?b`ERR)Asf zXL=9tXakHY%vThd@|N@LQuAEVA+2Z1>Vc_^hiAmnN~JWeU~;|4pdt2n!Q$Xgaj6+1 z!oiM!H(#V@dvPnrlaiKy_9I~Q6%VBIUG)_3I%MRnZWxX54-tVDc{;zoy-6c2wm*)e zBbh^KrJw*BNMT3XCfhA?M|cYz(_Axbg1PG^=Cj@95QZJbq6fBpGta2B5l4RllC{ z7VD6+&bz4glMWR2MD4V{;sDHr5?RZ}U%s*c?Bc*|4Prg(!~- z)iV5<8yc1Om|lZ?Twh}`!-5_4E>NmNx^a4J*ZAI^e9Ac&85f1)=|@aLI4T+8aAf0! zK2_G;|8PxzJwygcV%K5h8d?_Z8a4tM?WXly9bR~^%IjZzG5QN>dOtI^Nnp-Yo62g& zsC@5kl3q{9$NppAn?SBkFp=gXKo0!`nO4;RmjQobtTlg@rku&oLDz~6TqOT$)I9@8 zd^$k2^nKV3Y=$q>(&!&l6E@xI#7p-4tC)r4(+1hdpznND?j>fvD$xu54DJBz!aY{`EEG309=Sq}Cyz1R;p1~X6bvZJh8t)71T@r3 z@$;DXy+=pfSO9%*%X{F9zT&82=tJDOi~~OQTc>mRL@E^Ny_)r*WAw3;>C$4E7lZF- zwy7ddto80qz8J6-^CBFmcAR-3oOOb;PtgBY2cqejF+qwwjk>Ie_FWNeAi+M7tu@ph zD?RJL(S3Q2|FqTaOLxnnk#Pq)aylRJZ~{Fu<)&|6|XCkDx-Mo5eh1sQ#0Eju!vqEkxm)%V>ZT*P`jnqf9Em` zuy3?~2K!ZGt;i_3RN)3z9Sd69cAZrla(O~VF?1`6J}7m3r9`zcSK_`P-!_~^W{7CP ztazxg-Opf0GR7zFx5i%d&2hqH%bwyf)<(GJ3J2!j7z*0_$eedwl$@N;I~GejxX5d6 z&q^K+cvCm}YYyKJ*1_V&d(ZQWHum=9_iMnu>CHF2-nXV6>l?AZOA)aMO#2zEs0@6$ zW4FaU{tvS^1ABH#@W@da?D8eVEy#EKL}GUpq>{_&7cd)FB=*urDrNfD&BbwbL+{~ss=_yP0c|_wE)Uu6N?(OTl*OFPzVk>i%SS^rBh0tYmK#~` z!uVJci%@n0#}&T^;awZ0*}jnYumIf?n+Up5o`n$lTkIXR&$!ParX)${_mpg(syW@C z5?9&9wL{3M#2mWSx0uj=Bb52;m(BYv$>_Ix*xH?66gVnU-!O~=e$@3RyigIn|0Y(1 zn3Kr(jsVpEtiBs%7*TaNEU_wf*2E=dEgu=tB)*aUZBuHmAlt~-n}kIE-gvHAs@r#(>Gj#nAZzz5F!caWLETH!*qXYR%roT-!N;R|IFQw1%Z8_ zz?q+K#&(B{hn;Pq!vSCYkx!XqGNE0&$qXXM@H>+`=?qU(mcHU}5q`o}?~5vM}w*iVjw zWg%(9Gm>)b(6ogCo!q!HF}EDwNEzjWhTpA-G*NBZvWX?g?4;Cx14c)HAj zH~}?!n#9@#SLOKi z+&4(#Tw&j$Tp#D*4XW3z?_$GccxH(_7bMA*j-aGMcUp1rxc6H0IwM`YxO>4qI7g8% z%BnzBo5Kwpb_(5&*T2uSs9hBPJ?+I##8qzt|qz6{T$+FK^L{r(K=2oDPR} zPr-H?TFCMeO>e zu0>FV_PZs>>(LHR1&&{U+vP#~w1EzZR$74RLO@Cn2`&N+$h3j>E3o9d#=+2hp_?7NgSgp4S zZ785cSS%I}a>YWo=R%r?s7kA~ZH3z${o-5E*3So>g7K1HWx7yyykM%{OHU7O$QL`2 z_qH!XaJKdG7fo&kEW{YF0P}Zq{@4WtgH@2oWrD^JWVz@8uyWp2xy8QEbGPQgJjjja(s!Q)cu}Kg*`5*CpCpF+ z{ZQQT{Ye1Py;VkZPz7oe!o;gT5rPcQ(psd2b&=-yp6agCBO-fxxX}I39vT~z9MnS@ zGO$AojjYq(>=JB*gSu29+geRG;^&C@%^DQ1h~dAWBlr&)hxI z9Kp_+si@Mt|8RcamLWiCf)FeN2-x4>jKWoK&@~_IY=D)2c?kR3xqU83wc%`why-J@ zImi?w{F!78PNwG)RO>g+F7A_x)&d{)yMK0lzL|+shIEFV`l6h5{IDrXGrsQEU)%^7 zEj$0Lvg>+lREzn@o1+3A-$r%2nU$=m9=MbJ;BtO<+wpvvHmG5y@#Y8#Uu4tZKh5;t zK!6D7^I_Hg{Rr0wgGt}ZxQD5}r{!}H2LyAuApo5#bJ-0du>EUlZz{<>UO4Nv<(%u^ z+@D+wEO3Ptpg<(Cc>bZKL?EoW3XAaB)I*oZ=WpWW$)rA{-Lb*7c)4f&IjO2&QjpGF zFZpud@7@t!GNnZz8kcYk?v=Ksew)LJ7RFE+&xJw_w8EaQGtpausR8sMEso-Ml5t$B zWA6qbeI5|6`Jc?W29Bk@$O_8Jncu&@baktZ#siZBc#tsgOaAudUG7GK_(I0f?BnA*^^%MmsD0Ty1jC4m9SiDpJ}T{| z_FZ`=FxH;-t|57wuo*|{Z2Ns01bAiPh+wswQhR9cV_yu+TE8r}CWa<9<~LlbP!`zS z3LRm!=3A&$M~PweyQYPfCV7}BsgKGk>p!CNZ$vUBaVt2FY-k3_SJZ3U?mSncLwbi4 z)a`zHfzF`uJy*&b6F2z6neNjAH9CZiCg9~{p-A!<73B>6^rNieJ+QZvO3H!A(F2O8 z4;JX9UT_U6kaLYXPjbClUQ!#MWY(p?zZldb!dGHh4dCR7g9$0@ub9Vn71`BD+I1HZ z>@3S=f&-bn!fR*V1Qi*8@uSX|5NX&bl2N9*Vi>z+k8d%rpL|GYN0+An1cEPbEkie_#R4i5512$5x5@8&II5EIIXo!O zZf*{v_SDo-|4>u9m{*Sk;#Bbs(l3d z%ocOYccCPU`@!$m`7pX=--x9(Iqh^<0R%!YJx6yzC?NEFciDCW(N9S=d1WzSG|vTY zP}}wy{B)r&5L?u9Iw2VlE-j6n<#PkZ>~x3_%Xy91yUD?JvD z{-Sp(i#OYuT6ott1?|YU=Z5a0%B84Lk01{xHw98z>Fd17@sEDuqnEYSBhlKk-K-Yu^#*Vp@XuuZE=ljR5;9?8KGvhwqR`Vvv9z zyYPl-j^XJ^mU3BS5S|x|H|KQmr0O@5Uj6ymygb#?=V|x;lb6}+NjI-)#VH;raYYO9 zb|k8%M1T$)=?w@XcYUvUxl#nK#4Pa(7yTXCDkvsb-1cidhpxnjq!!O9Z}*hs@VNv@ z(~m6>XPAX3Gkq_V74YH47H9mop(xtqwl9oE{a~zS(^* zwhX-fv4ZdP)u864NI}oTke#-+(Ko%9a;UvCl*|{|FgH%Z7AeqkXiEP1NOYh;IAc#@yl*nY#ju*RGFml{xEx#Z zW3{!~F<*(_a&So6%$v>QbnyBqXxCQax2fe~*qfMtn2yr7IwvKtxNUh_TRIE$JMfjs zvBCQ!e+(}BV3ojvxuf~G@aub0_#0&G_3!@jY-|KzYTTT8Qr{D~|6PWftSs_8%RdRu^{o9Md-;^Moe)L(5dv~w!bhh@f2MU}9 ze`LIH8jzt!SI;mlB8!1A)Nw6DKVJRbh>kDrDZv1Z*!)#}OF@MJroQCULgxQPwgdqcNX=?E z`&{x(1qNo=rbk=W;u??!Nw0)|YL=q#I3k}xS$?Jm&Vt&}5v|do%RM2o4N$;AuKL=^ zi0^ib4n$U2DwzltH9p}F=A_HLnYo|&alzM5+2aZ>sg0B@f*Ix8Oh)x^Rwa1htjvGD z1vVz&`#2}aFl7)m+Cobi9^EsYM|Oa#p5FVB;b5H&vL*rT9`wa6)EV7XF0c}2m});G zIJnS?;;DOfo4VtB?$M1bJOnr^#O(-(wS9P#xM^$Tc5ZF_VlfFpfz~4vDW4i7Wz4T?mRnrOT~WCePC$qe_^QL4+7OH``*t`G_ire^#hHH^@UCM zd-9JSMnFMzL+X>}3+@dgl0gR}1o8&*H zUaUz+!1Mco^)+=M#=#~gWpB3nCi9;W#<~DmEn+B^4!JfSGY=X>T1nWvcL1#8DEMm? zd!ZSK1vUGueujixj#4xqzFusK{OrFumEJ~fEr6TS<&kS?v$$sw!Y1#O(z~80`L(>8 zCRqo_rNfP=m~y?X|B;`vF_yPs!t3%SwvOD{Zp{tbDi${{CXsSwUHP3I{-C3oRZN#9ey%_VL#9$bDC{F$&|gI)ABygq8DPMNOb zaq8M#0Jifo5-2-n7Ke0U{{L7A+>{zX;T-L4_=^O9=J^LBHr|)luI6F0)l}bMR9H}N ztU%*&Y~MPiB#0JQ5WR9sjIPv@&V(yMbMj}xf8+2#VEcMedtb-crozAffQtm!<_V=K zL`Y#O(Kjg4L)2cO=Gbk$GR7D;Q7*?)a2k&brajnvma}%ecU%$pkTLTehAwf@_NEi! z2qARlHUpuOeuYOo0SoKeIYQZVg!IQ|MpfL?Li#D?+%G3-meg(eiiz$~hF~D(HIFEF zWA$~%1$wOjxfguq;?z~|J5wLOGidRB+K!QML>4wmv204bUTsVCaUZhNQeI$-uNINR zKPXx;WIXM`W09|+%G!T%jlV~S3M4)Ak zo%`7rr(N)(Gt*~nKe)MZ8U2K0EuaxIeT_ zB$8@yiMRP?dN1yPX?6j*IpEJ|dZYm>mSL9fpKBaf^U2Q zHDpUKep2!gYz#@>*C$W|P?Z-^JWWi$f+U3hZEIj|f|Pc`!fp(xq32FJ1c#zk?l%qk z#;do%g(=NHx%`DE-o_E;2uwctNfEKgT>r7qjteC$m8H!vy9TVjB_NNra2sU-L2iRi zp00*MC>UD)_fKX3EDQmLe6Y1!II0MKS)Jtd8sL|d=ze`>{Hj5U2Rv`m0YxQmYWQAc z@MoNe-^hpyDDnDb)o9NYTeJP;-^LS51n zPcD0L_RN`!Rzw)~{$hMdBWK|bE+!hxyU8X*@A^M^?gZR;oc=N1ST>;HU8>HKk@B8KO*3*`2|Xyv7CjPz2u0bq;d^7q6JQ&I z&;I|FIfCg9Sd>MZjazbR{vFoD(l7>0p6$hF0oPN1O?LMsA>I0p7qDkT^%{_g6KzMd zjcHGNx*}BzPuULU_xUNTHPUaOzyb4T3EgZ+td1BOIo9$?&jG|zun3Kp`JPzFV0ko@ z?*WnRxRH(HO*XY`4Zg>hWD9WQyU)skNE?I46b;#dncs9xSGIWC_iAxw0WcnCpE@=K znT+yhbcp6A-NDf7oYNP*naf*dL0YFz1o^FzOB~+Xz}-QhJ~gOLvEY3nUl0Z`3@N*I z*@E@>?HJwZ>wjk^H8JBoIpYQN72aB9y2b3h!|lB0A6J8r5hHNSEl^6)Y`h3a+a-Or z;*Lejy!$4MUaaY39Nmmkv>2IYM{2)gFGc=LXc}aC809;Iq|x$NS<1sIm)~O|X+^-T zu=7!ga$vd8Obc`3{8Le6whypQ2a)*3+Y62QjD~tCCFC^GZAUXd_zq!klk96m+KgEL z^bZjzA%04A3OEk_Y|{9Flp19Kfy%R{?;3)}z;a&#uQhZ~fj#aZ+SuRz{4WLpml>e3 z_g2}nQbc=B?Qe|)bu#c52d7$73xVJT!gJ4e@k`Ppbf1upZpNuX;unzm@8l&TKk-ls z;#@5BieXAcex#FMbfpv{N>u>TU})Di?psRoHC`^~6%k+%9X9PV?_us3KD}z)oYHGU zE}RLKt>DM`mL+{XgUOd`DKM1amo6-gbRM`a;z-$5`trs9js`LW$dMOg8<<1HD+AT9 z%g`7IZ+=jb_^y7BNqY_d;?ql@UB!Cxc67UE7p1lJ=#Qr6H=~Vb~uuH}+(fhD@q>G}6 zqaeKMNSC`}p4wvvm2HP5PrF*tb>P$No=7UK2!qd`k?O z0r_K`JXzVBdaoD6#l5kLM*EMlAqYdk!e~eK;1|-Vtij)qbrs`!-(a!GsVye;Ti7Pg zW~CvU@n+yB^os!?>*2kPOsJmDp{JuYZGZCb^RW$17m_8o#ZRoinEd=)T~d9TTnEyRQ%1* zyKcZ@rL^miYSJdB)(M&%(eNSBcoNBz?MQlJ-wqml5pTCRtv?X&HNwVt4ljcS8J}qsHqQ1} zBn6k!vL6U8^1yl5O>TzKXR;qZzeoB;ZS9B!CxFpZtof|LrNjY{+Z}G6~Mv9*h?DP7UgvypE99t%4}9HCDY;VPoJc-_=_vJ=5kDB7OEx zISBdVbr~~smB)MMo!vZZ4e1hJs=IWU#R%PgU4H&4yM_?-UafRKDO27{u$3ne@qrDu zH`1KKJTLcaY@9(@_;t(ch}@+T;l|NE>(3b^l4bB zy;TUH7ZT+Wv|@++sL_AJu+-$LL`b_EC+OBllGpfEi5=-gF}1@J(dI<{toAh3J({xf zX-JzQt?&#sLF@sO!YjLb85k@?V0|1&D14Km&qXZaDUO5fIko>pM~a|wa(*5~$@kcH z@h2@p^Y;h7{|+nkxRr&o{FcD3;1ED^+@tvv1xM= zU;VXcj;)c9%VF-m0>u%n;b?{HBnOK5BDTw(IxCAcN@Td zCb`iwn5ic8r_dGyR;NH7Nq|YZ8PJV+BQk$9pkw!aA1GfPDxBG*sdRp6H6t`YzOkJ< zJhDf)c=#aqzp=ppErfx#(r;e9X<)TFHB#Zc$N>$PihG5dS%ZF)Ed<3zuD0&MwEo@q?b&kgE4O>+c^EU{$kEU z&#^51&!wMoSglU8yxfqB_=K;B`Pkc@$6o#-;Za3(`8^@cPld&ERLD}yGtfuVC}p^b zp5;f;XH6>_f?hy-q=A5p0F&V$nw{M52qu3uz?dbM$>BaX@m?$lW9sclonZ1y{dC6X zM=n?(i@&j83~)XTVm({olkJR3m7rD6tS+0PXy4+*$8KGEC!Rgmd6@I|y8l6O=T$)O zqQiyj487;qqt}uZRKiXKd$#yzda2>54itvp;AeLt1cN^#5}{2A%L2&}4GOL9r~GUG zR-^^g?CZ;qE-f8ycNHH}i3WSJwcHMB*L`GKLBUW-3{BHNn`yGtW+Yr$c1nkHVZA{P zw2x5nTeFx&_wrujLa^nV{{O-i|N4w&0QB?y2viW=N#T!EkRUdk{bt|KnJw1Afza33Nmpx%CZt}ezW_JjQyV~ISAu5zP9|MVK5HIRo|A3lQrBS0#BFjsZ?dp3j}P6 zN?CU7??3JqjzVfL(2{-bN6cX(ocu^I`M}d_cYmSQgW#+YN893!aOs#0gcjyQ*h2@P z!IfML-3LEeh;a4ZYyGQbo+a7jijbV9nG5KDR4ulr%JwhJi!*IWNT1|rxEBV=qV2}z6H_nc5(v(?l^!&skxmM;H6tz@rq2A@DmM4JSC^~@GGH5~() zV~i5ww7WZ_#Y~s!2pCfVOs&z-=(X9$+qEwJ^_-+;hiA7Fw*g2~4b_27a)`Ko(R+%w z6f#X{p^~>9ZKlB(eLCwYVE-apTp`8o2gRn&vf21#a2M;-VFZK@q;zD^5 zO7dCI=p8tI$P)6Ue9#ML*@~%SP)%3K@Zps3!Eh*q(2+rqU$A<3WmfBl3*4oi=dmjj zrB1^Btd%gjAJ4S^%oL3}=*Iq7<|4tNqOtrQT2XrqaatEIUAN;P-|lPy&&hQc=T$4v zvV({Y0y~gCY$JWJwbI?6FuIpf?)3qB6aXH+0IuW~xX8NaqDKGfDklL=DqnapxX`0V z>x+T8j@N96q6Y4z_6XIFidkU7ir1H9VGzfdSLVhWnruX(c?KSQ2swL-s&dSS-HNC^ zqOQb3)#6?TSG9rhij!{8Q`q}{201RW3}3Zpw0GGBUHKx>rRG7d+I>-toebp9-~=;!fg&Tf8VyES|7(4Nf$dpezs463l#;7D%)b>IjKIVG zXSBMvh6wwArf57Av5PQqW+l$21lc?LxZGvm&yee>nnL;uvEG|xE1EvfgSkEDE+a|X z_;qtkZnXOAK{Ap9i;Ns(w>vOs<;oZ2%0g?I_+rz5z5+4mdi=t!2OUBEj+a;vwypW< zu%P0#qq-MlAWnn?WRFfg>`%JNF~+oC0#@KteBA#&I|(SmR=(wm|QU<(`fwfmg+&j%oahj2HNediHCBG+I^9OI>_2-jajroT7gH7MYnqjL?99s z9*NfyC>BDv*7UoNl9 zDDTxvE25-fNBQqiog+Oui!xx4$)jI=zQ>VT^Z0WivjsO%D#j`Ua=QL?uMwmY6O!KH zlYQAD2$rgA>t7vY1JDO7M>%T9!ZGza0Igi3{%OPB%1%SREzCR!O7Oq+-$3x8TM&W+ z*a_32(TuvRT3Ff-KE_e}QonUSO#b5;0i|+|ti!aKn6<0xCQWl0KpOcyFZ65hNEJVg zQW7CRo@CO6)A=R-#*r-IPEe_^RChLKYFdu9uDx3fFL(DHVg-E*ni7Q?B`QG72bXgn zM-W!;^QId_Y$(RGhXOLrxY-?*Oj+&}!7J^F1PDfElLm1j`*u@;;v^>f1$tk}{E1@# z9LcGpbrd12`LD_vnF_GY_qG%|ClX5O7+ts6B@fveMn+(3wVF@Dm-LGVTvbc63xzp)QnMcsH;5wk@(e;0 zoscC0&P12L2tZmUIS%7d#e2nOJCn;aq(R+Izmp06Z97 z_$^Y{SWy5Ub};>1nK$gER(sRjc;jUFNZ;qnZ1#IA)L9|8_adDCwr*d9!mE&OtSj%f zspC#|Ol6t(M^WK(o1ij!5N-;4%waI;pw+i)DDtQ;7YM2rINobtEi0FPD*x4nVblHb zK~EEtu)!xsg>MC~ZjQE={}eFPFA*dR!pap|%l=B&%B}6T|5ws^FLne}D11FJgsmDM zo^nmRo{aVWrU{qtl*zcsW7Y&G8wDx>R%QMd0-byLOB5@Cb?L18nE8UFu-9F;YA5~) zI0?PGIF8>Uxee=KChnY3q<)9u=9k=0)@;K-=5JGu=SWHOtPP$=w~$HD>X|h~vM#{0 zz_>?nC-bA_Zim*k7OA9DF38R93a}>co!qivUDN5+#03R(-|;vksd&x%Z36W%3%(o2Gn8%hhK4Hzzo7}U9*EKuEvpYQG? z4f~P!Un4wAQl#i(uAS7r!%FF#I$p|%bI6!b}FA1=_{m@J_SNAsI6yIN!^1k%?; zI!3cJxG|8|QXqj~)6UkJhBt&GCdJ_A1T7~+ieM2d22HJh>>B*Do`@nO6QT=nj!3gVGnles+QN!? zvya7G6Xio`X|rno%0%c`ZsJ!k(~(FU6HF$Oy{3oTr1qvb1^wV zc4&HAy0szI#kJ)=r}qtFBf;vh6y1LufhP78XixnE%^bh#UDUzVwn6x~2F3reDBy3( zK_;CbNxKL_E=fn&Q*@1Y$mHC_ji(O#N?Od1t_R;s5cPW{OZ&8iC~F!f=%hS9yzqGV zvF<6r4z3<_OY*uNcP6Vl*(7_$ut0yONYIr+@GmEVmQ8M#Y@w-kr7j>{!rDfs+?+yPKCC-{?q@ zVJ`WbCj!KV?h;L${t3!rSeO@&zVIolmT*eAxH(JxUr`~J1*l+P7MJ?%6+LX`6}H1L zF{w)h*DpRj*ch4QTrfhx943m7ukcL8Lpl+q7j@a=4mcFZarM=a>u_i>8G=DH4gf!EdL* zzc>ULPA+c7oePMIbBE*7n>eadFVeLqJ zPxcn-@D<|6BjOMTp#jKSvO&TXRUKp4NK$j_2BLTsdAkCavc_Y-RCc$cQ4E`z<7kg7 zhI*L9j;%fqYvlvgfj`0_1pXlip_qGE?XZbc0Oifjh%i`uyqmiirBw{9(Q zZ?va%#G!7N`6Lc;H#ZSh=^646f5-!>FIr67AoHG%ZvPc-A6A< zZ9z1`H?>A9`t+1;*B0P4<6%OIqdWB=WTB*~2<2$!Zz4*QU z6MR(M|9U%AKA-cax^Q^bH$F~de*(C+nY5_lqCjwNUveOZBuK9I81xdEQkZRYrkLb^ zctO-}86IB7+{v)q-Ghbnx02s4GjA)~uD1*S?sWVyl7GP(Y6_sj6d^H3*>b%bhl~0i z?aL3TL*b6Yug`r|m5pPtM%#0CAEiSQY%6@ui)YEiVrp9Zf9@JUl0VN&1cmpu?g(Ke zh~%*6QtwPqfqCvo?gl3MkSF@XfX?*q;0^&p90+yxFE#tK?r*=Vf^@C0*BWu{!BCNp zD_I0IS5$lQxU}k_@O+O0jFgQcZTQrU&vmd$XPrsM;XF5=r@@OiJ}nnf!CEUc3OYZL zJlXtr86bP-y;Iu*`7T|fz2u*os}M)u5JpD#{TUcSN}_-`m8V#t z+g@Bw#NNl`_hB9NK9HF(=Fv}WzBt6bQF?dIL-nn}7pN(jYotHmU0M%bgk_xrh*szz{rv(WTCBJ;o-&nX+rkVu?JibdR9at za`pOG!^y`y`4C}U^vobd%D_m|S*<|(=)*v&Yz@%1CQ9>5tLH*kSrNptTPkm&B@o&p zz90?eKe>2X3?U*8vNMGFi4ToLSd_p4L~>olpAZ7sd*$n&XYd6gVnl>OqxJb-Jfioi z)^M*H8?AaJH~91A?r@KiVLi+-%6#;D$(Kbxf_3CfQ8Xzf#ZsK4a;Hb#bw^`E-e3ZZ zyCYFExH|Y222X}b_sX>o7DLmsxKNj_cKQOY!aWH`3@QKBq!q$i#Ye^gv`MKo z9u-L=8<0?QW^if?pa-fvVwGKUjX1I@iDa#07)AXB}@M?lR$#dO?uiy8GV?#jl@Q0o{}~gqN;l||JIoRqVp`Ck`^fvDM@ekGvGiOl=Dpek|@Ms9;;Q%VaY*(i+K2# zo4p3pk7ZT0COlHXC)U>?qLH(YQHK3sMwF~Y9fB4W=Y&ugEfl9Lo_55Z`e2P4#z8XX zGQwJrS8i9yFwz>ZaXVjF|8sgbI3DK-f}TP#&wTZ#9t)OQ%v|6Tv~XT-d+f{LOw&== z|BoTw^XrdK#fTnCxseUi_2NqQX3!^lWK%qV9@XD{Ww}r|A4|XPKmKe^j*9-l6ayQ7 z^LvB{Vw*zv_<(O_y#zRUXD!-995=peV?H|fQ96^cDgBQ;!q;WG=%p*|c_7_yGIEH0 z@JbaV_>U6VWa}-!{?C!?ijgEQ4mzj@u6R}%a)y%+L-o3M`kV?A?Z?2_IaG^=eac zn^7pCUS(zvWv$oV3^YIx_xd3=ce41uFG6KXh|PNkId=rf*mP3W#t%U3zhxk0y!qM$YATf@*Qr$YpfylfrAUFM?m_#| zw}~j~V=u<7B*_QW8XtVMo?ir5=;2r%5C@1=K#fi_(1UKC!*0DO$TR>GQ&qFUGfkyC z$sx)*m(|<3|KYi+!-EDN`UK-;HTwbVjXMj>9rhz2Q1p0VxzqoA=O0#1t%qs98j7=y zKOm!gRveXv#!X?ZcUxaYH72;`_Pi3Yi8wusgMZQeE`VPiQCfJVj%w2nfRYd!WB`8d zZ)dY@hg%51niUrfwxd4p7BILg_Kl?J4iOv5hhW_t(KBoc9StG@H_jilSkY(5pUn41 zot`=!`43uKnfM;PT8xhW?yCzU0Y@65&BKK+YaZDS`s#dMjTysBeYNQ2S?ite&hz4m zX-QqPb)Z@1?>Ggvr(^e!wYF5DK|w~jG~4;rF!v>EeR0&`VOKh))CZCz=cHuG1|M@m z32`@n)7_C$gq{^VwvCK#qiI0jny~fx9!dT8GMh6ZKgrB7MV4g*9bL+?I6!|3TVPk& z7xVdI`!zTH>IT&|FAf;HU41?%_$|+y_ySgDP(*J~xV>juq-RMv3S{v1t2Mmu3^B(e zxNHtmHOWDC8O2J-&6n(&RvAG|7yJ{Dp?K zk*-0?-(wN0CurRASUk;MQRhqNS}<&(Nl0!!F$j`;A5pt$0B)Y372cei?<~~l7aX?w0*%XK!-5GogRiKdoQkTtCJ=gv2eli)74lL+_>AV_}w2sL3<7kKnTAUW7zX z{vX`63@+rUhOqI9N7#CzrRelb*N z+F!~PYK@rWgd*@JX?8!91og*6oLCFa0k8L1hK`BX6pxF|US%H(gWQsAI}Y%s;*;IY zmRd?5b|5Ms&TCYS^FQ_vku{zUy#6FPItGi+q#K1>$Rphh4P!PTwsj7vdQmq=@7|m< z8hea65S!k}o%!M&PZrEZ&xh3@3+BU_ZT~0sqi(@qUSHGoigf_D*$;7iSB>FdfeP)j zCM}VUeDNjwS{MQHe%=;frWydV?vt6n;s1Y(y>(brTi7?O2r43qq=1AVT`HxtMGf5@ zBHdjAj)jOw$^b(N3@IJM&?1Pm3`h+O4Gu6g49&Ok2zZ|3^L+0=Ui*?gYp-?Z?~b)r zwEIt^yMofLVi>p3C_m5JEd8g_&w?@r9#->vEi4D@Bu#BkfT`}B?reBcG(^@9HH%)o z>p|V=KY!|4v166qXTE@+MADaN=!ZM4LSECkp)NZ%^#Y)}P! z7j)W9=KmZ?n1r;}ze~AUiv#vP!&bSw>`~nD#MdUdRJjN9|LiUL!!djl2Y(h5czuBt zN1QSLt+~g7Zm6mDE~vX_mHS{t<3p8jIL9|5xlg4PKki53I_yXD0(xEREa@W;qSx@U zA)bB8ADWmVQHWY7U1cK=a31xn$~;%=2-29NM-&=YzZ8@4$fZalkgX>@Y5PU`Im$fe zny<>I%$U*oKU!`L?mZy=4%;z=$0f~47~@@b4YKqw_SuYA;cWDB*EEVbzx4S+9-ji0 zhE?7pAU(LMHhcfr9u~zugZa#Dp0*70yKEPFW*ks`9OXw1mvGId&M}kE&t#RYd@BPx z4Y_cM2h~Jk>-Gzwo$cZF%8WT67uU4+e{SqJLByx|@QQz48`mUZAE|v>U_5uvsq2ISsO#y?i0(77i}zRIz%@X|~`E$DM%-d4%RSN!+G`3FY)Tynlr3Dj#&a zani|ED~!8E4i}djCseD2us{WGE)CVj9`6))@{Q3zfBl51iFf z>oM5&oI;dJ{Z!0-&c){$_IBf2&Qdd>r^$>PZ~K-!o$8VraozIr7Cwk;KZ|F4`$s~z zh5H#BZFe8_;sc&_dp1ex%G}kfv9nlPB5nMXe(g_44N|XFs!^A~*F~KJyYo)BbGTLU zJg5pAxi@?oqL_ZJU0-UtpKo@b=jA@-=Z`&=PyNoR#ExfJ5hXDXLR}p)MmF}|FLSw~ z1DSEHOUV}uuD-tT1MyPyYTiWY#>(<{x-31#pEki+i}kh+7S#z4@iNTCUnS!KY0b?1 zciTWU`6yTJeBN&Si752dO25_q?)d|yAm_Gvda01BFCKX=pEIhxwQv^6@zmwDy+@LG zO_RkzLW!ixhuWF!TQ9d34CY0Gm*HlbufYy3&n)^HGcsw8hyC*b!B!9+c)P0e{kaB& zBTMajZe+)tLQdgWVJ)?;H31RXRjU=7`X(IEiIr{&fUT#T56kThX7=X5WVSMQ{Di9W znf zLg#u^haXokcj(ajPqcYdwsd)mpT4Ni-;wEG)aEmwHZL z*?Ueh#Sy;WYXXxtxXQ_$UdI_;R`1Tel0;~hV2i$bOUNLLB;Q_WtI=TIkAsK_s3ASy z%+CDd^bC@~y2WcZE8eOm+R?SzX~o0bY^`gZ$NRY?PxVz}@_<}?H;~EOLocj8@~LoQ z?fx7uSA6bVwL;I6t$Ch!r-6v@(*JP`&M_ScI*Fwz^JY4@ zhen=q(A!lez2-G;E5Qq|yYBpH>T{ip@KRwBhI%~aQ5a`0+7R=5H>UTp z-3)@BzHv?Kim<*qh#ct5sEhzAo zs-ZsRlwMzJbMYn$onSP&;J6@)yg-?l!)d1t^nsiDK5`d7UgN8{XH&ER?^I*yM?;H4 zD9wdl8zM=((lyyXWrW*6C>NVGD}-f*BkD<;}8JP2$T<1oD8 z;1c;iHZT4*6pS(WcF!h)O+4DWxPq^tF{uTv^uqI|>q6YV?8~>(mbwaiqx0q*(o3Om zPOV%O?ugaP1c?(3)wA(y(V6WAXM3Uv6^k^Em=hd=*TbVpRlIvXzC(a}9fL=JcbVX5&CJJrL()uIWhYh8UJbR?0INx@_812+iE>LLug}u2F*Pd zll!R!QXExuMu7yN@VTY3VkLC`V&3Hk_3_`Lx8sT39B43I9=Gv@TvwKYNBFmg&YLX9 z^~-4S5DoQR^RgB*$y#E!=t)9tk_17icpD7t%=CHohAt!*k9;{=svoW~?N zb)|urk%QTY)m`mqd5loPn{h%)K1?;E%~umF`6pv+d_pfYlhLk_ISIbf4EIo@8-*S` z@==3h*`Ch+D$iw!hU%T11pF?`%AE1Ym2kEb_3p&!((!Mgmmvd%cI)pSW< z8`(@eD?HiMqBaR{wtw#NcC)S_UNA99;%$E>Z@&57YUa~vz)Lu->^lTlc*oeB?sMfx zN_Pl-QmGMZUP58Ms61LPjE|(a&XszapllWJq#H2RrV}YRrL)ARvI^e3!Rad05~;xN zH`TQ~xBX*i^2^ZCSpZ%WRP>TJDoXHcXB=l4u%JdGtK{}dSdu8740hsI`@<24J4SQB z1Ab`T>U;#VnCy}Wg88i+>f=STmk8Br{b8z170#9qOC3zoD&CtQr3P9eyW|4S=no$k z(##@N*5X{epC&&2`Q_cWqHlg>RTq+cmJxn%4euiF#RIhbW#I%1~c?-kK2G)tmTa? z5Ck=yms_6aROI3}xdf3vSauqoSUZEBi`)moVoH8N<5x{?KM2787lmr{!ogR91QAUTg^G@%) zlystKQ-Y1_K!B9{+xU2)V%o(k)G2|twz_5_y+e12H|-^ET0UY|9C@Gs)o}?#V|040 z#;O{~u_(o;%YrvK9${g{)WOe6C4pu%8`owSUW`o3g+!pXMQ}iqdG6Tc16^$<1A3#q zp3mT0Uv5NeTJ#T}ez)Sbeyh-i#mGYP!p_pUvTheqlSw-^mywk;I2{d!(4GO#a1(7N z9a5S8@10CXn1+dj9Ph&K9=XlXv z<{o2#M`SSpZlAF9nI)eE=?!14L}!Z^KBip}uJ-p?TJKuTeY=fEhTK*jK>0g9PygL05*du1; z^2|7-XOqv1aC7vsr#^IBwO+jqd)b$@?F={R7XQdA@xs0{ckeBxlbI2*JcftOszrW@ z(P^gSr1#+Nu=lNVG&fZx6E+94-h2ZO(;s=iFy6IuO49w^#-5}v?)1#V<$dQ}&>|B% zla3TQaIMMP3qte6kGM1Yxi~pD?vp*@PJUMX(ZtcDiMO9bLj(UIK2eI=!$;X}`<=TT zJH|qlh}K~i#BgBiFX^Z8K;)?9BWD;*1?-YKRRR zJw@jXk$rlZ{s?4YyWl5Gh&E29^9A<6jMfPXZ_^)R%_2?G>h}eS%upHmqKzP#0uJeG zpD!aP2g-V;@i6KrC0^bxn-45vhzWk(3#1P(m)H=q&P@qd*0XJF)?BHLgCbn04jot{(R2XV9r;zk-Chg?kigB7=&VJN9n6soWiPG;U|w zcwo5x=2rrrVk&D2FT08opO>)qQO@#yG3xj5=mO)Tfuv4>oi6O!wEGCeEAANQaNEiS zoMn{=xuY&p_5qFoX~}j5@C;ivUp21CgQ;E`(wlXEGfA4{qI&jybe_N9ro4VXZcNxS zY`^TDzq_ptvQ7_ElJPv0hkIMy5lUh6qW=B29*I_mr5pU3rC3h6!d-h$ZWaeT2cgnJ zNnhU|eWd3NNlmlIEgVIcBo>W?Bf~*4v*oEh-fY%6F~gSba#EN3caR~DRN^}ZQ*YHA zeU&S!{Fb9LUAb=A*&QshL0}$|mFYXi#Zgm_c6diQK{`8_d^IhNx7|dfNkbRqc$tN? z7r8+5dal@8Y-1Vbe%|tmD7_JSUPSYrg1?chMGl=JM=jnBi4P-vi9o~QuiJH`sEKw> z17?hGtNC;|+@R1BD(PhKq-S;?fEA?PF6Luj&Wi9eF%rMSzh0>)RE**3`D-`0|W^WDn}tqAhQ%>`!b~3Bs0r8@aHJQIj)RQS%><*ojU0 z9(BMUz+j9%M;Hz1mv5bNYAxi^ zgDIYN`|j_+^}Ytr8ZzG0J`-U0O}rGA(OxWiF);--)GrMi@~v`Pd;f?5+RQv8`_bNU zqi+-as!V#fci+pInPDIhxUc$>L~DqEGqVzK3H3j+dvvuVaMAJ_{s{t?qCh5c6TsdgJl%kar)8Q0>SgDLvx(V`yG_m9KI& zU%_XpX3_$iDE!}nWW+koULJ8PFIks<#Q$&_t>32wZby4h4Q zwPz4v=s#fRmle@CbWsPm`?u6Su9j^%1Nvcs#<$?5LCS^BR5n1$){OY|-u+(S6>_I; z$E1U*wV#%+gwcK~qVQZdlxmredq{m~doMZ~E*ClDW$UQZq&2cI6Y}B7+|Q3lF7@Y~ z2DZhDh<)`FNjrK1TV3dfufh6O1URg%V-KZj*c$`2Q0`_<2B+BwTU^f-V#<wHe=X3-_Oogz&MEXNeh-}p5y4Z3f8va zXg^V*7JZ3MTE&1?7IDO6M)G*gxlG%fDnx+g)>K39hQGZt#A(*TCVOk7b8sViivEY4 zpd!w(=+T2L1j|!|*_;j?i!Yn#2(K-LD#@FIQ$fmC{Fkjne&0R%@WJQ<=+rmGrXB21 zJ*qO-Ps)1<-4-%j~~NV9bBhZ2j92S@kt9ue*| zem&!;g=Coo_xYu@n_uKk~POnQPA;h+FxNBVQHs(YXh~1ChDR30Ee%+38 zimlq(>YaT#w%^b3hBpIjS6aw0Kc{oo<{Xr;@y`1ZT@gQc1naroe9nC`GjJvMEm$Q> zWg}7DZFsX zg7rxOi5ucNrr4`9wyn4IU(azuyQY^3k))pA0$|Zo7wjz~{q0PPhn~0$WbSSUA9SuPZ>|lTHVS1qb05{FBg8!-;gBHe&xYCJPwC=< zKQZ}Yl!e+}vqlC**XdjjK9~y-?zB*=XdF2UHyE12?k&49XYKp7|Fmum6q1EU+gz z5_Go`pLrKuri;GpR~JA>g&4b;)p`B%&YR}yeUGy}CxtuwBz zQJUH8A$xLqkplM&D$~^>x4nbJdErego8?K8`y75E{El)Zmry{}E7wYN=*zHK9&nZ> zU-0zV7zdLVicmdqUmZUdM3OCpJMy;g$C4o82x3mhBtE(-`|pWuP0#mv>}s9SE}k7# zH(*~s6b)c@EhX#^LX(~x; zDA_N{N{jsl4C`wUZ$G)3g#unAwGf?`b4#|MS}Q8eIsiYZbV(L(@YQsPgXg$|G$_UowF$TSO$6XMX#jtxrO8MNT_Y4hcQkgv-G zSbWWQi!Bl2xNes`HEn=M>f?gy3)k=%9;6>!fV+j|iGOEMW+RPKnCh4tRPsr76%l+g zclC?;*9^NcHOKddM7SaV`R%R2n-|JeX*}PB`-E^-gf4uUqu4mz&SO85o^#1{V4#;f zX|odk%E2N#t55EO{X?~isb1Wt+Wd12)vK&XZEb=@=C?-#IMJSm+I8)lPnai5fcuCG zTJpq&Kz;d-*I8L_A!rSoQ4~dPST8M9a>m+>n&Sw)xSWWqr{lZa`3vQzKT?H8?Hf&3 zzXv_u-nOx}t(1%la(Oo;T_ez}-yJR#*_|`vV>ab(+oM4%YYXb2LvcDzXB&u?f3DKU zK`i#MwJVOo{hu`4_m)$b0PMa{P?g&18q+i7IO9xN6!7#fs$$M{(>*UYWWO}z>!wB= zqoBaL}SjsR51xJ#oYC6%OkA$f8zZmVCchZaMisUL<G$vE2nB{>3R^WwKY9Zb;_MN%>B?tDzy>( zHuQ{2c%G>06p#_OQK(08=SAKz>m;6FPa|E1>aPnyr3IIrx`$6Srrpj@Dp5^HGeM^B zoHE7@ri}(bKiJicunn*F#$1+o&i>o|oG_h{1UT)N(_nFMV)gdyyAZq8HjxiWJ(CeJ zLl+gdWd`~F7HdM=vA0UeC2emy$k>_Z2*A% ztnnzh_8Wk?aF1HJ50-l_^KS1_3(;h1&Xx`ZYwItB`fqSy(wg@AhKHq--zhJD?<3<(>k)UUu6?uU8jn^s{aQkk4(i=0t>x9J3WT>=Eqd_X|Y4_nK?GG^nukyYn#dfW4(Z{gz3!&xDwa^1$*kF7ij+Usve zeOf#$;L^RBtIuhr$-(Cd!#2r3QU#~~HosB>7uMjPy{n~ReP@LSC$iLG@>gmhB^}bG z8Vcc~Ytq`70_<~f2XE!sWuGMpDntKD#TcWz>;4;mu;U0vE|YKms947agtD;wOi?6C zC7xFup76TN7#-)}Aj47XGRwhHt?aC4@ka zP#L+_%IPpA1cLY!E;_Ws%f{~ohrGbCNXvAFT`Cv8rM6UMN0wywvS@?X>S8C!o6JLy z;x}d`^jk_MjYg9KupUVF&waTjU-CClEz$w~mM>F1OP+kZO5Eh*R^_%){CUrRS^(3~ zi}o|5ha?uprl@@g(gx+yWDM$;EjPBp*5|)ub1`>By)PZ8_!~~1Ca^1sogSEA^WC)% zg%Mtc=jVwVFsZM6Y5%D)6z+jT-3{n}K`L}wxVr-CEM8cG*4Vynf~jmU^P)EdHWa0C z87;#DmhIK-;dm@8Zg^bwcU#04qgczUemonun04XzpwP-j*(r1NZie0&Aib7U^029h zW5H9-)Ay6M5M(C%{1X*G3NJoZqxik*24j;b;%{?H1@NR7p=J2732E&^ewi`Rd7Yrk zel$Y}+RKsehdzI$>5USqy}lW<`;*wiF6qGvydKBz9fwgBYF!-Tsqa(Ld--1ro5pvu z`H~;U-j}<#+3?JGX6zR$Z$PG_b&MqRXRXd~aV(Sfcag@Y>Iho`agdrVeMwW5 zm7wU?Tti#F<^_*@9`Cc%!l-V1D3(JyvbkwfliR=VYkSln@p2N*LL2iPyX(v}MtW65 zNd$+eN4b4a(@N2ZBH8jM+m*>6KWnLki2nOfOkA3z{PZ26#hUh|#%IpLd=!q@-n$W_UlYz4$= zN~N31s`mhd)HSl^89LCk$LjnxemjlFR_HS-wS8{jIiwmU-)Q(5sZXqb7uYx}W1UpV z?^?L!k?;3KFE%S@vvl4vV_to?`_36~*abknkyNiOj&FK4CHGq62buV!d-3>=DPYRA z8b#UqWDeFKCFYU`n)#wC6^Iv-O8EaFD9M||I?>)I(4v=W>3AS}i}gY69cY0zKV||1 zP;Hqcrc&0Xr)dOIBP1FIz5a%j4xAqlsW)O&v0+q}FOaE@nPuM^2*GOTtE)djeHaSk z|H2Aj$9|(LgD;MR?4^MT(k|=Avo)HnL4;Z$wp+C9(|bLHw2gD|S}Kbq3xxJkahGS{ z4VeMyBQ~Fe5Gm{?!uJG*=r?l3VQu(%KAIJNB?l`>Zjgi=wo}y zlgzJW+%znDcq6;KSb>5k_6xtTN^=L#p6-v_x2yA}>LEb96Ap0omd`EBg(I$pfxz$# zIYWEJRbxu8d>=0&%%@?P5|4eGEWu*UO^0d$Ig#nE(dhP?F6qa)R8^sW^Hk#C#@@+~ z8wD6k_PbP@ejiIP>z^Tp!Z+?H32fs9@I?y!99wLX%Y|hquQmled|-kV!R#xi1xu%o zp!q!_U%Sg4Y_`#3EWAV->7VFW^nMtuKg^z7H6j>1vbsP{U^E}5MgVfj)@EL`jH=*o zC4~dH&v9A06y`0wHSzF-Xh%i>3I7n!Y@k!`06E z(`h6Pl!dj>=`yD(vw!CdUwGNPrS)sBT|cvEzFN6!Os!h-krp6LE|k<*9Fnyx%2ZRo zw_yIQKC+HQ++NE=akS<96J203+-KCqk3T83NiftNm^5V4^m1if*SmF{<)znO5M_gG zo%yIjL6em)cc^M@{lEt$A?CK$$ZGo>)kksY?&wmBh+y?aeuxgeSzvaaMwYCI zX{lKmiT2#ekw|&9w(Y%t=L-I9wt{NSA82}7i#&Oo`-}}huJYAp<45ZgwhU0+nXty( z@-*iAc&#XRt#^y?bmCvau|g^*TjxEKQ0t8t)6i$72yS_fnmV7pO^8U2*_j3ZSu<E_N~#g4QCQydDw_c7G(c5LhQnLwz(NZ*uu@)Cej_mSNf>&3-4ms=_H2&v^rm; zLEkSm1vNYvWq``&qTTmvu6;M~$%g{w3=Gt-%;ybypCla6yJIaWd{s!;lMa)%oqY*L zzmv9r8H%YpgxJ^p-y-(XUN7)7AEzXb{R*HlV72o-5T6Jr1QB^HvOVg^2@DF+WN_LiXyJmb+7sB(A- zlWAE#q$B!0`j&$@OcD?lsR5#}O^uEG(>y2CU_S$f*F;|O{q8&VwoG8#QCK#lclGPQ ziF}{DDwTW(Z^BaE{)`1ht(&p6Q^t5tL!m0gopm`(Ga*4PgJ)KWO!ss9#9wQAA`Psm z?T#AJ$(kDQmHZ8Swz(WrQI0Gjh;OuJME{i!fIH(CGzxA&Ce>2k3Zz0x@Q@m1`v!E!gC$?YA5skbD)eKmNE00_>Gs^s zLErJ+UsMuMPUICG#8(Tl1DUw}z<_jT&;5Ur-5IR!gMyz`Uf$}u?9DGOm`MjQ>beYw z1EayA^3AO$3!W(7T@o*+Pj34}Xd=zr8|_%OA32zi;#Yq%=;&Q3RlHzr*-Es7obswV z!`<5264TtCoq;3&*+|8>oV95Y4?g}ai8FZktekP%ipO|NH1b<<)vdB4i+{Y@LklA- zGg`d}d$FMG_XL%W%sAaT0+V+sKP*SBBC)R{o z3h9gV$8u@r{rul(QGv&H5t5knQ*y9LRB;HN*sqrwqsp}2eaVno zXYtDs4k7>?_%HT46om@BRZ0+GQvG#RoTQuRV31DU{%Pvp%YLGDD3LyDv?pYEL*U>A znzap`yuIp^C09MkjymGhF$WGE8@aN{lnrK&8o4H!ro;nlN6cu5NxlCE!X(E40zpCc z?#qdNNZ}~32`SMvZVsse@y@-zA=CdCwDSiW&sPFuG!1xV#c(n+Y5c^^`yyD50L}b( zX1db91G+y{lX2)!E_c>Hx}U6onnVY#WFsgAutQ%r=PR-yr}}>ppW}g2t^-zGz=%yR z=$HU17r+oUPy+q3_EG}}lA{tZmVeCs_?IXDSASTi+%#s^d-KN_r=jz5m1thSbtsV2 zT~DK~PUHMg<@|SHJ|6RLv$+qx#~`a+H7~h^o=79YtuzJwC2l>T!TE;|0IpkADV2~O zD@-yzo^7xE2R>Xq%an zg=+BXJ?Hi-;PO4_8fJwmKg zf}d|wTdziM+cm4FEK39KgSeiYdNS`@hjTp3$7Fa+K$0aQgS~74rxHp3)hkH%bAN%3 zZ;I;d2DWv!T$mv!yR^S59Q%H-#4Pc(*V)BX$EKJ9hICiO-|K2WCMf z2jWP$tGewwgse@i=Iq}5AN5j32%!2?YEawb2=V+AMH7sCbNhyXU(Eq2P_{NK3jJF| zxQpjXB}Ww!CG9^kGQ-pv&zjILa$?;N=i|ZwFoU-gL3z9`sb2gF2{v&7-tukt2hS(| zTkU2ntt;qED^6Tq^elcf>{-=Qnh?NnJ&dLVxt{bH+`yiW! zVHUJr6aS#DKgX9^#s~b|doS7T$I%B>t9HE^=R|k@W$3=W0*o|5Wt!$haCEQeR4R!q z8CkboM$Za=GA|5HqoeWu+vv^slsu42eo{2T-2waB+VY}~v!P^vvJeHZ&|E)e`^1S9 zvEKAg@9}s8_kql>&E7>)wyjORGX0-`39NR3ds92x5!Av4NneNY@oZ)Jw*z?T1Q0WZ z86Haf90>cz#K|{Tw5{RyBA{DYE{`zJ#bNt&X(B(0yi~LQr>9E3M)VHKQ3|4YIeAB9 zfw9`|O!%Mqvq?3;x&lpcC)f)7j!5k8XXy`mcs+piz|vS$`EZ`t%-_5L-vHw>P{>sr z3xWOW7+IR@;J>}fw`hRaz*Y;RKg8z5-28rGOvl?pM+0ZtE)IE?k)z8lDy)1VIU(w| zx4;;>rZBbtbBr3ioIZKq!3fW!&J~=qVV#=Se(Il+r{V)pPu@|M_=I}@eUd_u#fZt+ zctdMZiw2BzB_aP(Hn0MKx}&UBXa7<*Jy}#skkl*C9^G(Z_}aRr8tm=9-@}tXl6*@8 z78Bc(#`}LQCixm+n<2O0EOv1zl}*laVd3907I`RRYLOLx411y~M~rb{y*YUXQv|6_ zaL`lUu7PNjMhJJDX#Xj^SaQIsMaK1E^1V{@DfLEDO!7fwO9>XJ1{r z7H-(8tuMJPSIe^|wE4LMa4!Ewv;R+PWRx2XWkITwrL{&{JZh1&v-8Xvrrgr++V#fC zAX)gfQ8hMU8i&{68w_1Rc!INy1qs3hOw4|!9Z#`=HohJ6OFm}-heRQ$Hmw)ws7 ztCH){Wd*Rs6obJL9>2}h+H;Of3S-&0?#2LwW}Mp(xjFL+^DfI3hIL$r|C|7&Rmc6?^|R*!bB0k&KK;Mu48c3tB)YDvL?%EspW|0?H*^k=q5RL7k{&)4`$$h{yD6p%Y zJ9Zr>AmZpLiWVGH3&Qy^iHjKNO{DvBrj$3N0|=W47rim8w%uzXJ$T*JB5gxo&Wx8c zCF0He{+gwAr)YJE``!U8zsE}a--#A^F2Gw^6Wl#&rTH7_9`hBrgb28#xP=0(caw>3 zZ$WE!^9|9d&i9d>n)$NuP>xK~pFzk0UvfK{Jzb!5+w59Z{nN9&XU^o`A>{wl$GDc7 zNHO3-@6fx!$hm0;CEy4h?TVIaGrM-^c6c{1RHo*91x>bvhYwVY#<((DZ6Q}7{t#FYHh(8g*Q+O;CKs=h} z5+4ZgYf{Ig%f)(E0W&uOqh5u{e5DdFL4~AE6Q^Nmzp{w!3zTnQf3bP8=zj$oz+Hp3sK96VKuM7)DvH z+y1+$NF%YdTkAkVuXHv=s`)Gu<15m1r?shei2-3}B1TrdE6xH0nC90DdeeVEOBo%v zzK7UA9ubwZ9?G&*-eG`RdZ+Kuo+HVPSV*aq<@{u*Z<`YVZaIcKiTk}txP;M7I0 zljcJT5AFSv`^j^z$OjXX#MGgS%M`i6s^ zTv_Ys8|SuPtL?=h1!zkP3Zd=IuX)7;@A<5XF5V5&$6aG{M{(V%cm3=9T9WFDJ_t>G zXGV&SWc1-WMCHzi%0v4=i(cn%7*#jYd!J)YL6^hSLzFZX~HPW$9S^x z#h~ILU~!C*cn5h!4}Wy2)xrUJtri?{is_jl`E63bUVw)p5A6##Is~Qk zV78htCEpgZ?E_-rxk%_<(BHEi-dAF3A68hsT-$A^_ch3fb8D_COEdJ%N^CDz;;5F& zL8ZyLR;|0)vx$En_$Sx)EvilpI*`BdV}HD2eP^90_ZqCK}r2-D=9|Ic30XA}!-5Xi% z-Pi&mv-YFBO9~IL+!A9+z0kwuvV5OWgr@Jw%+$!4)U&H~c0zUyW>_^3oOoj+KGAb@& z{!v|E4S%xO9&BqI>EJsOunF6de3JLU>ryl0Nsti!p~bT}QKrIOm+BJl>LW;+a@fxo zWh_2;*v}UwHTvcpOt@w>>CUVeW{L72cLGS3Jl9f(z@e>+puqIG`OsSTH@~X^8Nd6} zb+Xsu-IEgxq5=BDJ>Gpj6DV`BzH-X0oVlZfoS3AAVAv4pTjpX1?ZWeHRJ1hrtr`TIS=f9oE4@&Z^n`GY@iGFd`L*=mFe; zF4AP;_4>R+ADKC?5#puK(3{BV|o6WcQ+T zkbPc~60|=5+QZPdCofvKbyTF%`v9YWc|ncz?N&FoCW5YG*#6_7?to!J{Bqcw7d(SN zp+`hHTUXF4fIM{(H3K`dxLUUB3jI;tK8@DVfg6l#!CH57p+Fh7rzrL(2QVL@Lq~~g z$$49|fo|u%Wy1%w^{BnNe`#b)3xh&Ek*PCxf!cIAa&|fHfjC%uAL|}q0090 z%b8yar%DNgs4e;Q_A;lykBN4zY}G@d9E!V!aSpos#Z>Y@P=k-y{K@q z=E<FGUY=-*0YgC!agZ|26Af;&nQwe7FqTB9pjxwlL*J?sRn6#tY)F z3fzrynX+P`07FznPQy!vuUsam4Xudd(vGSaxTx&eujc+?-n$0`q&&oRgc}%`e<2(a z#X}#pRncVwAcp&OJkZqz9P&((d_D@2o?{o3pzy)y9BA)Sr-r?`e&=SaM}sgCQ)eKU3z>}bHYeoGZawkENi zU(JTl#RG&D(C99pX2-Fee_b5#82I2?h@3;*+!Zo@=zNq@(1EyLa&b?op^HEq>R!N- zqlSK*ceO^Uyui-GoL@o8`1MiJSlabMb56UBuahf{6v#Xwdy#+DtXm*b2cBt0r}-<5 zO+cJL8|g0hVPjq4j$5<&kK`(9-G~WOG&;Q&NcF;Bg(f}b-(Q&hC_wd=VUSYbl{Xo6cp$He$wyGM z_x_(igvFMRk8IJ@NT}6r8h<~g;|4`ZFb$eM^X?sO<_U_|FK_Ip%72aOHbXhKydzq{7 zsWYOT)uDA;^%s3lu(A|_+^P7cJMDGxHEW+?&p81G0ErDLA+MkycEi(`1EM;%953|C zC`D`x+^)8j8O!wK#$dwE?eh55r^iXW%1Yn6`47uncyF%I?ubjY1q@h6r&YE~A0vcc zUnF^xBsC%oqeXAvY&$~A48n=+-svM2BddIqgZ2iSEqu2{o)?QFB=A`z)+KZ*mX{jn zSAZ_s+dKswoS(d`XZ0qw>Erq*-U~rDbJR(hkunfjU*XlS{lA(E-lPLwsO#yll_1`f z-Y7FjY7n=ucAlm;JnycQi#lsxe(YgD%#QnPSqX(L9xUPfhe-n7{Z3DqUsw4O9cx%K z=sWeLEE4_mACdp1kiY&}CeJh{lx)@V-tjQt!Vu^uayT>&=i8dL;uN7X*a0Bduo0PB ze{>c=>!7ch>cC+40>8gVH)^oSesn79V%1to!7HiP1c_Vnvn%DbNUU%NXrJTqNP?Ts zLm%&=zR_VK*ctf7Usj`L64Jin*6ORaD2$xjCk~XWwrd~(mBMD+e{dnOX1h)MLozmc zN3qwTUmq$e(?=Ciu10(1@BZ4~=JmTXWFWMcy?sOJcVH6BR}W{$QKNGw)}%!4$AJGQd5E zijLGR38m+(zTIfMlun!`+H^=K|Z{Lq`Ngz^AMd0 zsl5ad>s@sEoi38PgKu#AdqL-l#eKb^`!u&yvu!|C6*xznZ=q7DQwhI%u2bufhacJ) zrAK?$;4=g3zv9_jSN`cB+3)y7eXSE5$|k`ojw^-NswOR$-BZSN!q5_aNW(!_qF=%C zw|J?1RkZ`py>*}}&VrXP5o*b`@{z?=!gO&A8z*mx4bRO<%fn>`!uo2~7XtRgFmEW2 z5|dqrx$C%m+Fa>m1Hy}2-_sravzB-rRl0jECHnTQrXRSBsXUr@4^rkm%~@zyQ6tc8 z;g|h{x#gdV$Jmt^o75+jAz?%8qbV&>wm{;G!d*jPBfpHAad_pE5R4!z|g4&U0v5AF(e_i!l7rdoQT!sl=cA|8%y#8(x#^9MZT>B^Mf zE&b|U2LeUF1g?Kb9X=mL2G+N`asM)HxQfnQe>U1Zx3k4!ybDuU_e7qI9>g7O^WSd5bZP{8gB z7Tl=WMBczC(Tb_&WUp2SQ08%6ejCvSODtbt<5`GlAC~J;2cE z_GGB7>R@q(XQMAZM`+UcF8k?Eiyx8e8_B=a4RYh;4}K(ZPZ-{iwl)hz%*)ZLO8?>#yXa}iQb{Gx*ivULi|q@j z{8PI+_S%RaT4k09)lOiy0eq~uHEmedY6++)C6@6DdT1;W&D!k}`|EHf|U$geWB8&DR1Z~L&4CWNB zG-KxcOD&TDPFjSU=2$y3S*=qNa-MDt+@7*rRQsAGnuOxV74v=cuToeFGq9JTVnfh+ z^d}O&WxKOi(aj3WXS*mEf78GRZBZ@H?H~kc!UkUk2p`1<)(~C|R{J@~>loA)cO^ma zW9EM0Ke=>pIU&BEwfMLKreAA$7SkC3YIj(}VPM6uRL`}hR8qpY!q(CR0E1RNFX4ZS zcW)7jkWFzrgMt^}3P<+kay{a!J!?%C#}J&A`bB*+#Z}tA>(?;f=#KM1hjnHFV`04z zDLVqK-0*B8RWDqkgrZ26`&d%}d=k&09sAX9cgDC^Vlnvo0sTL)b$~SHwv5OQ^+8B% z^4$6z*$QA2#vYBg?|~7k{r4}vPY#&v2z5nyFPZO?S{Kj3C^p~^{IiU7w}%}*{m74H z7-Iu+frg{agvc!A=sXBeuO6^~Lh>$k{4Rom=K;V1J}|Mwh=DBLt0UIMF%^A#;@db6 z=vJ@9>ci3!#Dln{9QeB5rErHkZNGrXZGhn4T=}P1Ah)RGuyFbJ79PG|pIXoDzv@HC zs9PIcvSf7>ZpahvMy0FH$OXce{_8&q$qNX1+-Av;>c(e19g~1fJDo**&b9=LY;kwN z$+U8L9?~Y}9WlRhJ@xp#{ol$@r^0F+<+=Vps{S%As`vXIhhbySr4n zOS;CQ8wP2lyKAT+1w^_V=^na<_#gEB`QE?#**uunoH^IIuD#b#Gl)EgBr*M1nc z$4O(wMW36`-PQWJKh&#H#+Cd(iWC1;eE7B~iVHAaXp@xoE%F%P#kbxq7~KDulXpyq zM@U)Q9$}MZx}>dHfaTboeQya7J+7#~)R);1%{5^Rrr9p36X|`I<(r|J?;6Dzk&mLbB}0! z%D&^%#j45$ju@M;j;MEB|67YR@d&ou_4DUUmA>y`u#4~!IDDH|;TJD*bxw1@j{#Na z$M-|oXTIZ_g`IB~d7Ea6N#;rS^-7vW-s6Csu$uY%ARxc-r98t0ZfE!umEOu2giM}n z*7W?0mStP-Gp-7m_V&w6n2r~I69zxB6DKKXUvX`LUw z$GWt^o6LF+2K>ZWn-D_>!|wGbc~qD=B%T{=h5f5&2)605CcB^Y9S+Q#CJbp;HK%)1 zXqyaTAN2Ok2`H=mfhsmn6Ef_p=Kq(nzDdkHLZRU3j)E4Y<1#FPOClWyB(Ko2yy=E#lOcLCa|d`6a7_iT~bW zm$ab~Nw)In6IA>|EJI=S#?94FP8WVTG`wRPg53jI%N824S}VM+l12W6|4Mt*HXw-r z1eIwub&F*+O?NKpLR3>5Vbg(-XrqW55x!3I_|hHsp6Kd}1Mcg0!PL<-;5iuuotB-_ zy=P?O!lUSXn~gV@`85+ti^Yv|8n*V^$A-l9^BdlfwRl5Ab?eo}KDD||))$sBE1sX)1ao3!FHharN2qCk5Au&iy+znEY+Xe zI7Q+=P_~~V0Vww$wQ$7Bgb((4J6cgMyXH==dF&@kQvM3&)W#sVQsR8LyF;Il$Lb*& zqG^`HG0_3b8!jt+p`s8uo2mUj$p?@bCP5M*Dh{T-`N(oS<_nXthW8i>kF3(;pDph^ zx38|=v#hR>tW~_!vT+T9o&9Eq1N}SVjA+-b3=5`FT`t}!%@Q9&Dh-A2_nuluwJ);S zPpKx8R@|gr6#TY}NRQ!LegUoHnam62LoYw5wug6||+P#%{P7^SdgLieHK$^Ht z@ssd1c4zI$ZgEZ;FV=@JYa(Ew@I}Cy@Y*X8XR0qj>E9o^Kf^POk@>Q<>=is^*$?t2 z;E;k{LU#b*8>pfo#WC1zrawey z$Y&5}fTa&7xM&)=jeNvvenxtCn&<|@?Q*MDg)&(0&Wg+@Yn6OGr@0+sXb&iFIB&{* z0&7+;QfDdP>-zt5-e-=3B(kjgW5Ne~_e*-YlZxZVwpvP(mG$A7QS5dwcPnU_@?z=L<&Lsxl@Bm;b!nF9^Nf;@wUvw7Xf^6 zA`w;-d+5UZ0p$!txeWn35QQ4UY+#ZC0)DfZ8`N`^;$DEe=z>xTFcqNxhA`vRuADt^ zCLTT1t&(T$9!Z#&D!tUF2@fbm<>8Ic<=)*Knp|3|O7Ke($<}dsEj&j-87r-HRv(0i z`mgFH?u?qAYAPo%B+E9HBWjzpFkkjZ8e1)t_}LBE&1O#Q<$xhNgz!v-WJj2 z5zr+1t}m%?eOlYAdYGYSwc>W)ux}vm6kfEv$X=JkV55VxnApDsP&C){1QL!5lZjjF zMSDks#vu#k0^=E2#(&y~GXXywx;#T9}*&YG4&ka&O{OM9zYcAHufeX=f zIG{H2Fnv!iG&D7h+wcpS`?l3T-AyuV&oEA3fou6dW$&Zq;c~{!tkL@l1^maZdiF3F zUbBu1_dQeM&C*eyGXMfI%>Zzh@H?U!fM6&f2`XQsT4x~9A7NX!*xu-t+Dpb643>{Y z{z;$N=@H8~f9mE65AbDNSWj0>3eIzWdusZsWW9^`7~jF$Zju&C_N5BXyo=>I)ifG6St@HZ|!Nm&|t1HH!-&(Ms~# z&!@ix>c;okx4U|dW6=);lpCGOyRthF^M2MXdL{t~U{03z#hbqh32@w_a0TI(pNEM0 zx0~cTo9LxoPK3Aoe7cfi$xP$+k0(7w-{D=CA6o{b+}1;YP9btop*EeIosp&};?7Pk zV1WIK3p(%FFh1AHlmqNdiwCP~I9)C9@gXes-?_FM}&Phy)-J_5@B(Jj04*T;cI3?&RwEz@ zSo!?k7YE22fDq=8G??WuocX4D2OOtYsL*B1FS^w)9Q!A5;)ph8wIup_7uNX1E#I0G z7&&OHFWL9$-!7gr`8A$Po+&IfBHbZi6b^UX{q%P%|P9;1h&}VUw3x#+M9q+L4CnC*PKnGzQtN( zdvjL*_PWDl#1EqCDj_M&ab*CH zP)uQ}C11Xe6|&(o8T6O-KIL{?QeM#Vz9Igfdt)5*mBXiTZUb^^HfHPC6*NB{XuUUc z_ISJBjK(t%GzCUt3Z6+U5WTy-1wiPo5>YlTs zpH(gE2NMlGynW_hyIqbb7aW3tF?AO^2d-%dUm2`&@uDxaHzWaO86Fo75ZIHu3^uz% zbyo!!7Z?>$7$TtU76xct;L(S!GVixGKqer>Y<*C*_|ktSHO8LlJtvdFhr$nXwZClYkASbDkiajw7GsS>wXq< zCmETpz$gr|#Hm$m=x)FI>)&BE)5tt=+Ws2y)#Oiu_IbUErFEzNS1KKdj|MKjl0%$= z)!m?lf{iYdm&#?`XN$FuSF1$jb>OUs&M9O4FXH5AKkwN0Kjyi1mtjta~6dP%s^z4tJG#y%&+nja8)J98w;!t+#OL6i4;qXzKukujoPi>2yejoXVw>|7?fUKZDEQ*a zd7kjOeaiLfyGX@kj#|=X$3J5Avu@cF|Ko3HASn)-gyIo%YswPwF2&|ycIRTt_A z+XiM)5qz%|pKT0iS7#|4Y=M4u^YYndRRFp47p^C=omZQu3T-%cpX4i`&rLcA-a%et zbHnNt_TekZS|{}#Cu1CFPN%BYlL^ccfbEp*e$NPLSj?W{DrOYCA53NV0=&O>P*DH%FQO_ru6WRTx{$h z;PFwt;}A!VO5L72=N2Eq<@#n#iwh7DScCTKK-@!Oz6t4UX zPh7lRoJ+rRKgq=4!)s=?OHcs6$BS#`E4eZUe~eh+l$@W?waXEC>`iRWcaU#svj_{Vmi#*rAyBJUj$j z+^YQ>V9pyHbqrP{vSN1$7s6YHd<7l2WxZz5legQWiP+_pVYi$Czx?xqS{jm=J)jE~ zdC8lO+n!A(jXp&G5|s+I)j5wxZ3e>Zo^r@HrEqfqe$&4X-NidFgZm&FEoa!yc_LVi zcgkV?oag@DML?<&1|2L7O0%@GHQ##zHR{f>$(6db8YYfy$4?g}(z|j8#)?l2&LaU@ z#PHDGxaOV;^p@q~H}qEBz?FU-mz$Y@$|(&)<=?v4+mJYV2Aq3gKO=~JfP zLQ&Twf>zN&UPeO=_l zWP4V(1ml~5|LJlWe>}>zp(K$hj3a#X)ooUz5N&7r6^6r?KMV-ZD1mde?QS`*mwZj_ z0`@J4)y9M!bm(q5nbuA#eee1xez)`8%~Qmim)qP4K7If-v02RI=W%8S_}N3{m+*2X zHoMTJ^103PlQhXG`r?ojj0)+VA>hhiMWt?-(NhVNWx|>9hTX_9xuV# zh2L8iNalNh{sSYyKv0&VE>Yjk4(rYNf+AxW8F}k%X}etXf#51TflCx)QgarpET0{^ z7SD8hRB!Ke-OekUMraJqgl8MtF2Is1A-^Hp0p-wmT`TTN`{~YB9HncY^*#+dq{%X< z9oEzvpy?d&X2N-LWiTb*v_(j|)5XRqjEf=V>s$1kasTqexqe%HqCDzqNe}zjx$PO$oGli5D*b#4JQ8DHo<#X4Ht>hE%w%YDH2Rt zh4vQ(k=6+p&=(pRPyKIuEC>%$c0wUJHWvf0ZjFdKR={>)%8%TA-T zg8|3+aC-S}G}sYa|Ge!;eF|TN={99Scdo=y!sjGIsqV`z1?jru5fS*9Ms@tDmGZk_ z4lbv@R`qWf7eCRb7wkUr8GV-LbQs4l;#-(53Q!HeQJieURF}6M|@C?TEN#Dw^$`3r=Vrvve8RnCzs}a&QLP&ICcP* zM0wu`}s&8W*QpJ6DTEE(r#sNKd3lI*CIpWT( zdwav-(Z67CJVeE${#{&a4tf_5HL$-4Ik&Qkvb&qGn9z9hP%mIMAtC)LeuH-2s&L)4 z5GBt#bN88V+vWL^8x{3D)kL59I>m@_hDW&HpXva0h+hlo=u$oX#a~-XrY{cpS;gsW zeLH%F5##-q3umnhjsdz>LqA^3obiupL$d7h-5zgg~30!z7qSuXRVufcz%)*!0~6y;o1E`>xBw4 z-B8uYJO>d-_@Uj8mTTFe!}opE_Bp0^Poyu#Hb5}v6APjO98mbvRm!tSwSbO?nI(1J zx%En3lu!lHU*^xv17P;HOH=mmu=>@KV%ARd07#%;^amR@?#r5LVYXW#DdvT<=34&9^VIDc6j-Ed2kg4x}*?c6qrJ8gp*A$#* zg8cF#NA(ka@2U#TQBlZ$?h5p4Ko$W7wMhO+v4XscZ zY_zKN`t044TI@f?-?!j7B+o2P!j$z*r{p+i?#R~7Hc&(j}7H?j!LPOmsIa{K~%P7cZSV2>3 zMQH1P1Zzv<&c#~Yt1!656M}S7yYi0@S;u5h-!L!dr5x~f%)Pj!r{N}-SaB!NCt51UDchy}n{ffm(de1pW93phzqs{8*qz-x zzw;!f?NeAtn;M+Vgp(gYnqXY>OV2}ci)#_!=z5{b7GW7+E=G!8)wQEs_08IpPI1v| zSinrA&2L8qO@Gmg+S?X{u9I#4MSLN^ukovlKfUO!+kf zW15e#=Ee3EBEwBLRIW`TrV?WiARXAxI7av`8@j1=n_Fvsd95vWN3`*k1foFpPs>J3VUOwIIS9_KQ%rNc)%Z zkdciWTgpqTWAb|*-8a7-3G;e8S94C%PwESx3`l+-xT&9b$Ti-RUY0hR_H_W9w?0&l5xV7vJ(1(W(R|&71*1bNT?UFsn$heDFN8< zn9!LHxoP_IT7E<83INN_5^_y2CB*@r*QJKYJ(wgf^9NoRFn&!qAJ%NO`R9%&$nme{ z_Rg{r4-gZBVd7#bf>4#V@jgGiT9$-;f(b&m)%JnoO$!8$>(z#b0n{%O_SN}1Z4XXC zfXaULlkh+zVdC?N^mHEZvJ5A1f;*l2N$Y5H_uZ;!!!*Pb>q>i-@f9oh!j zNK#X`xv#0w5=7Xhut7caK;ncsf?U9%n-{b|p|*?xK7 zU{C)(ed+Yoi|rFtL=!Kvv@i=%uS#sKP;QR2YGI^dz#$LUFR*&T`ykS(wbgw_XuU z_nt21zs!njo?WLDv{y(Ldg4fXi15$cQ8h&v=lj8M-+7;BDYwW~XS8F@3M(A%V`~)V zHsvB8&Re$8NPiV9W_=M9DITKOvgIJB-B%_CUh%6QM1e21;5HO!FDsg?Mv2SuvZT~| ztuEEneOE_~zl>MEtaX*5t|rih;S{`>4hLtDEE}jGS@`af_c(ZKJ!M%QM`+Ozy+{g1 zQ*e;}CVVPeA-%)~3Js)<*2C=m>=ojZ_MuGt)4*)Ia!bbRs%OLRZw&OSc)eUl|OH9!xP5tqs8fV_XBT*h&YuLhzIMJhHqMl$+Ri zQJA77En)*?3D3Mq+V5}idM5^~Z!McmCZ{xCE1Lqp!wjN)V=83t6^h|eqe+Y`MQx7R z>#ALB8R%ieo4~{B`%9E`UzecQ3YDI$M}XdpQ-A-9MAnt8Ce;%-?F&zm7i%K1ga|4uo|N68wnb^lDWclnDd$%ux- z7r{9A?NvD#s^xDgex|^cV0vMVAWQA@ycU}T+GrwRxLb!*Q%@>~EH<~CIRx9w8Eg0; zw7DBA=F7cUCaS1=4TVe;voQ^afq_F>3QOJ_O5Nn;H`bi0#-w&MS=1VG1yH!lXJ|*zqVZTzBn;vo;8O2-yfsZvNc?5mdY38eu0EFpgjSZ+Q@ zaBtJPc|_MI7H%`|@ak~%mFs{38ECcaRP<_pf(u-n;Su>~x$`b6b)P(8>h;P(e@M9e zEOlmAe$5s0N9xeP0D>3xr2ST$B3PcQ7LHN7Z=?+Z;6bWnUt?Mwf-jgfe`w2yHu(s? zaF?)ZDN_m@N&LEVstANOVzAg#+h8 zK5X4O(eYmEeK6b3x1CdVA58N1$zhJNHCGmNFDyhWTvER7A2EQlar9iNL*~6)X7*~! zb-KR=Zi~LIK^wlmnVt=?w5OrOMt-TA-h^%yAYJh@03Pti{Ge?L+8Y`P{~B_p|LFLS=O}6j)73aG>N)`Ov_WU%*#w&Uie} zxZS$gNLi`Bw_-+jICvp$t4wsqc*USAnBK=+uY(Nrt8jl}=aJX4bjGCGT<$ua9}`sd zWoUAwL?pC2f!`=;?Oa;XRdzIvsujJs4n=F*$|fp?VNRC*O9V1PV=DVgzpG}PWjo3x}QB41#Dh?9| z#u!r6_#-dLvJ!%Ap@-9rhx(=a&d&Rhl*|vzOJXoeP~e!HH(^kaDT_A#cQ0F;&6M`@ zEg#^Uz3ixK>EppOEa&8y6drI_a7^iqggaKP@X26N%+?7TmYBO&qHWPrsmi?TuAa8E z^11T5!Kmeau^azOImO5JDu&2z4tx1(mlKo8uT$;h5^BVOJZ*jSa)+82`JV2Oly9`m zcTo885v7e^ER;P#UjvoCa&xsQ@TZoBwch({UZqyE4<@{q|DsZ_B&yk}0dL#>@f6)1 zUVe)mg=!SeeEUkiPb&*;_Luqt$tbRbX?)K#*BUm%4XetUQ*Mn_${u1U>yeb{m17$V zp1_>qN7>WsW;{+kGMlL)u=slR+mHUjO`G=1V7xR)|E&~Y6p#J^M@F&ysxP<&%G+yV z*37pAG9C~%Nx(CEch%#`nJsLf=aJ%6gFcis13tf=)J}p_SX2HiZyi1X5)uQZ-7HFK z-{t_1P<7)|nPM4Ev;&9O+%bJu2e zzxrL=D*##+IC%N-JE6V{MoTzletitp+9+0=!O}NMKHU zd6N0@>p|SL8EWd-U;4BnTFfN>7hB}QjUgKkN^l*)^RbKb_gMP9&zacW|+RR^>?o=YRT&g=iugqH0Vu^$|fV&Ufc5Z7H z0&)t38MVx!lZSAkp)N)80TfAA)Up9vA2TX2$EbIy({S@U9r!{h=0>yKFx{>+=dN${ zRr_)OHyQTt`$DoL{*IeZs;`m^RmTc-BK~HAq`abgo*NSDcB&)Z#8xJ|nlfJ5&nV|q_xy7Jz(QmT3ReL{z!)1yR?qp!j8ZU10mILj zJixw{9M67Dj3=T2gnmt0YN)Mr-YC+(0YR@PYI=_A4n?_n`|ysW?c1w@8(;4av?ZG7_fifyN^?{5lK z0_XAtdMXnOwXmUzjA^x64`47<+?(dbsb3zp_mlvmn7yx1Y5nV-Qxx|-9F8YLix7IQ zO9@3a?0#1Qvju!H-3(tcyy(N%VER@9*T5xrnFAPg0D`LsO%~HRq*o6z4PlA>n8q%B zSB2ZKZEpHF9H{AVEo5YrYhM;{UcQ@T>a1Dj~L{xI=8R}i^iT_ z{@eJ$@+HMxN8`Q6vXprw(YnRf-sk-_;zD5m&7+wM zkSlS-qi?c7Wx&0#h-YpZv~`9nFBD+YY1P+H(y5z%b5Bx2&GSVzM0^6uT0nC@HP>B% z=%B4swZjY5*6+~0>;!QQEP7IU6mT_VZhoVECy&zv%X)+LBvG%QqEP7P#;jYH|pXYi4{+~o|x7blzfJnG1)-v-7BO5 z?{E#cRK&JNvDrIs`H^Bp1{`D>W+Ty<7|+)JQ7P}KdV%2}nHn6f&z2}}J%iU)=Iy62 zIW@PlBG6>X?_l3>zLh)u8Rjf(#EV!ahi2GEve)^^lpA6(WpKU<%$Y*S`PcR<@Hr7& zyl$FI*~E}j?;o;cZR+(?KS=y<#Yw;udv{m=;&dp|+RPzR#&fd)jlb{#xtzK}2J^*L z%ln8~0vu4@S>p}^_=#IvJWsY0t zYNq#jGzNpmc}_?Ke@rUpnkS9&!SEYIHPBNgS2b5O^aGj&`K8)mnCyBusTcGWJ}|FG zxU48Z4op>FG)%X$|NJuCv_h~Zabp6ECm6Id`sX)C<3ySzauf@eAcg3AiqAm&Fe3tc ziE(2d5DkWDNbx)A$BGOT?bGw)el$j0Z1slW4;HIu3w$O}5ZyGS1zyqmQ*gRYvpGXz z#n=|bdB{v5txCKHJ^XGuxa-6)kMuyWX=?TR6!VtO#lHW65O9! zGqHqwj5hCd{{oRk0n=j(Bw5|p_a})cwIsF-;p0hrKJsy%=|^23HzfS+KkPM5OWmZ; zl{5JT^_-K%yY+e?w1Sn)CFLSI^B$u@DYB)lIic;^NDn=pmIjz)*S+e|DVP+k?&2I- z2REJMIW9BMSiT_*cX4Va)Oiy#@{f1_k8co`>St-snRiUI4Oq_B4xa3l@O__D%vmQ1 zx4IuJ9TZhBd>P@W@3S1*SXjfVajm_g(dauW?<=3EqTcfZGlvGGFom2?)*&<-B)IJB z4LOm`yNIzSr2XR7_cm4e7*C{$6sylR!c~*(07uDowd4mW5aU%*N6ElYdQl#3{EYLL zxCJ?&lVZjD*-(&K{gO)8*OYuDA~75aup}#~zjFa6(a@LS`LG{<#O`>Gw}uK(AGHq4 zTR(-?L4UeCEG9JY6)VcFydlvE!2=QQ%VmkqFgyt;XR8PG;8h!dG}#}QJYO3}=+UJB zGo4aoqOtb|Ad-{*Z;T~Acn_fW&uDF5dF%Wiy~alM{CxcK$ak-{s zLpI&Jsw4QE+zd5Mvp0n=?$TeK)q7GDNbe8#@^CQ62$=csr`ui^xzC7&js(+5lGK!Q zpv^YG7`UrbpYr4QCD__HMRdgL{S34Hk|)x0WDhnV`xdD%LetZT?nrLan3pVe zbNi(@9QpsOaVP$zH;U?}#Ac>Sd2jIkLYjAz#h{Azw`WbFraReC3y~F83J`NX|kF)cU*HNncr{#c+~ zh78|v$W(NWdIrmWWN?}P>snAo5Ptqs4;1Xps14RqB3zJm|D=18m(~ZbWkRt_RV|a`{F*aAAHXP*+UxTZ&8yIVynbY)6&|WWk;+CYL8I0vW ze{mdZ8quoq2=t0&uuW>UD==wPJK;aE3*hr;$TJnYM~oF8OxSgh{}L4PIi5c@uzNbE zc7DqUH~rVw5yngFat^OT?1C5Fz7TQ8f=H3{~ zRdJtr7X1Sz=m|3tgJ$__(8P*Q-hSNnCvo`EBYOVE>-pJN9I-b}%3m;P@)QXi2sh*< zcas+K$dEV3%EHmL1O%piW#RoBDuRd|$>ZPru~wsM^k#P|25EGN*pa#&MuAz#Sm+i{ zwKfvR&!9w9EI zLZloJ8*ZZ7>d8v2I42bJciCjRMrL~Nx3(}xpCi*m>rg(rIa8TrHn+Drpm=1c%AxS| zAZ@NOh-Kxt{jFl!)L2XHxLh9A`O7Q~`M*v)}qKlw#Q(yS12 z56n&uNw4WaZ@8qP4pZ^ldUvsBIa13~NV>6rl}fRW%9^ar;PR&3crdw11|&sTqqnuk z;Y@m@Yj^zYu-n^}Dz7gGnrHL(sL2MP@*e$Z`7M6J&!82ZV{{frzPuui+_}7B3i9_B~$I#HpHA-8Azjolz z`5V8JFahZ~we##8CBUF%aRK$2tKKV);Z|bsIL%=nYvxh5-tA^Tu3`ge0}Xn5NS_?= z_Q&OkbG`jvLLt%|EUSyRqMB5`by-xZajwkeA(AA1Rt^PPd>Pj7o)k?~U(yXHDM7*m zs%PJiKB$+YG*gcs!aOCW>~#|_5!l=i^H;xA##_VXKdBoPVsDoDaw(F#9SS(_6-8VY?~5oL;g~^Q1L;6hVAv?)00^JyD;OnzIhE*HPlfG zE$#d$zM7x{0>NPLN z(hNEOu^KmAF02~~T2uUHG_(J9`ViU#HEDU!h(Gy2A%DYWbRnj!TN+(ww8?G!`IgoJ zmy!K+?}b5)*jV1gR)1ib(&&W&D{OV;Id32iLx8g}vGPyXgV9Qf_q@{`XT|0rZ>bHF zS0v-^=vTUCKBRqU`jS$>-H?=AR<-<;O}@y_H!4omvq#Upw0@{~5u;?5z~%$4pwNwz z()90b#Mo9A_ z3)T@11p+VC{_*A`=}ggp!7e!#UHa=Pex(9t5WuAG8!Q-?qsJesN&@|F7Qp{+g~bUD z2|=%pn(d1Oq(Y{zb|S>*G@8vSXaAU~=37HMXTOxDvqHCU{ zoWI~9*3c&YwQ0JNF}n?qyalu6 zTPiw}`YV5FGG=mU`s5)M$k7kQ@VaEifu=E{X;ePuLs^WU{Q?GbxYsAqh_WeF)K!!Y zX{HDNz_;6C`3u1yDV`zd)L~=XNACm~PO%w)4gXP7O15K5Urg;L8X?)TQ}iqY}R z92xF#dEUM{xVc*%Xwt8>8>`Kkp8tx4bthp6x!iA^JUe6KT6y;$aSp5}i2mr&M|l(! z(PO~OMK}GZT0PF&2h}ExHrg9?K3Jtv_uR`vJS)RT%O(o7b!n|edDB{1aV4LC(O8Ev zb-6eP4FsU?uqZW*AAgDXIYD#~7nYys0g)@&R4R7!{Gef;{y$s_W=FMoyvcY3DLIja z4~i^?{;u{BtGM!DGvjvb3tGQPZu3c_t;N50ssD;9CSr!?L$!5ZMnonS(Qe(+WU)q` z30J&07Vpar_xz|6*Ot_p#|Cl;=ma16j}YwjB2$V%ayQepHHa~g1Jhyv=5h^W~+Zm z`$X>0{mA7MrK!~WftQB{F5&CeQWWeER!U9!ncd1f#F!4cis+cUAyTqd<%Ur_2aMz? zOdYy2^Soe{H_`J5RY0^e!5Lg19-aQca^|s* z+5ER+Mif7Qro199>qwyEV^8hT*ruWf09f#-JXBuxe{s`QNPq= zubVpupIN%W6^r|GoLGi)?}drAnp6xXl=)VkpSuli`w{R=&w;ukA%~n#AfYS2x)_;9 z!61h9*VF2b0fk<#koN>(vPIi(T7K)?aFgxN_QC|-gBa)zcR7BX zorc;!e%_Yw@iWi0F>up>m_%I>pVk@|AIb3bbs;roJpSU@gSso-6EqRE4N7$!G4Km zMGNM)i_KVq-DcbUdM#YL#AP1bLky~W3CdhDL#>uOt3JYROvMeiU#*WRXaO)|v_GcP z5h%{NDKiQQrsJ!p-QGOAywL=3F z37+)^kH3AMG)wRA`kFgJ;^RYoIQnCSg^J|TYO;iQ_sYnWiunLy5e0n!yH>oq^2-VV zjCnAQ3#K;=evR>#}TSV0)u_RZmocPmHtos7$UnepZWP$#u|at&gqllM>Q zzGU!W$4hrvq#NH&s<`linagY@)&~eR}@Zt+%dZ|wyE0S zK4DG_Qq8(V`U}Gi66n<0TK<@v@U^q;-m34KdZxkHR+*2>kAzn<#t6|tIt(Y#q=Waf!mTX? z89imEOTC-U$XAAQpkr%86vNrEfDdY6Pr!TBrnD}i6Of?+%WFr`__QORdr~ZO{&bn| z(ciF0zu@mO@YG65odR##Mh)4nx82~F)c;Y~y7M1KaRCZftz=e}Jan&NUamlK{2}d< zqJAM1%NQ!pD6>euDlcsrY5T~boast3buTz>xpvZlTqUtJ9-xCnr#?obtP9(`~5RJbcE&Kr_L-IU4L^6N^~ z-xxQd#*J}Z`McH zDzVk8YW!4V3-idSYJJAx}=+XD>K}} zqLCgeyumlqsdW3^57{{yQ^M=a= zBN6y)FMQAO1!|mp9e%WdZ&}%(!qReHQ;xdYe>WB^2Kp1MLRfe;q;>??Fwg|h?mdmu zZCIN^|C@b?=J!}a4Tf~g-W;8`20cV-T(!AfOk94XHes~Dd?vqumlJI=^Wn>kI~^uq z6lcTTq|S=Tfg+g~;$ePKfp&olFP?j@<7}$ZbM)ATAAIK|c$rG4`43eTz7>~cD|XjS zv0-}-Uf;9PW@V8}P)q6|DCFj__on2|P_tQddn`k1;y=@0co8R^buoV#yWU4Kc%bn- z%{)ok5(`638gkp^Gq`mHky*1@3_8aUd5c*og(>&((!)jCobxu7+yFa6{Q_igA^lBD zeamlw$wdSNoJ z>Yt#Z^YZdI`%h`P;>xe_b47dYm%bt#TBu*XuvimbxD07$2_>+nR@aEY{fEZ+Rb+nF zDiWA>`o-*Cl}abjwC$w*&_)QDpMK;5am@)U^3?cmZ5CT@e{LJCZPR*0X%;oR!P}uV zUlYG%@Pv!{kSyZ+&kr7i;RK&19#fhTd=QG##4nq+CzlNN??xn($p>u}Y{V35|IH=i zv@h%hhTUVEaQEfP{FVDA4v=!aLim9Z@p$;3QiDZ6+m*hNi>HoB@fn7Y!VNOlh=Neq z6x578iRlp>-Mcl59?z=)|Gg$e_E#$}(Sy%M6iQ~*J7mB>iTT=$Zou=)d#Wf%EiN1= z-!s8|No+!B@jnITJo{b?^7KU>TfMAo zNQ;(gUi9C|QAKTzY?E=jalCGrkq>##4_9_gqN-UX0i~JXI9lV}MZ!Eq5|y4o)ubM5 z8^^z6Q`BfXLVWqwg;NH@*+fnN#hhQN9bHfNc!*|&|l%nh( zDHZ9lz~48?+cG^y4rzFu*h5Hp^D?x_i2!Z)USzEWCgoq={|Kq><_~$#|3}nYI7In; zeZ#UWu!wYrbT>%1^pa8{jf&FU4Fb~LE!`o#vcZT84#k_i1Yi`&Cj`YAdL0bEl}+y1=}guD^dp6`EmTDn z+vmM<67p~G-4oVgN%aH)h{Tuxf&EWbhg^K2qci|~_C)!`3^u(9Q21_Cs(tlTPp=PX zGy=w7A{=)Ld9of2{=SOT^db5bOQSoA>RKm%owu0ntFvXUw_Tzlm6){oty_Qc4UCqEV@*80rIl1BZ z(9WgDBhM@LBdp78x{4FEYlKjGW^;NDQB;9OA;i`#l|q@`y?$x0T?c~kc%v*4($yJ_ zrPNNi;1EIL)q0mS?}M~4A|g1EG{JlB+?MM058Xjug_Q?6rU%AvWxnqPM($)0df_ED z#b`8n&DPZlVRX>?{Qm}yC9D|J=+*_PiicyvZ_WGmp3=*GuDNxRW2^{Zt$Q`=p)l`dL7bg9A| z)aYgro<0q$Ki7>7IN5i%GbJ15UptdT^(l*{SE?~<#H)p2@Wo{5q#Bt_Yx%U=Dt=A@ zpzyD*IPaBZylWAtz+eB;BgXDkd_{o({W!Ku+#gTr#uJn>YqHAw;;*kl_1pi{U{8a{ z!)pokM#JNbmNp7!s?>~8;yXsErM#_s*2^m-8R#ZXReMtXy?{`?U?%pT;jQpHC&Pij zdQWJymUZ#zxYyd}@S_dZt-cvYj}D}>T^`cfnBl$a&y8+eAe z$8^E(Cb7n`z%bqd?9q-SBBJ^yjn~aY)G1jk4Ex)aRb?0aop_Gq;|F(;QP~mvr6r>7 z(iXt7^=?soOFQ%hPyNkxJqzv(?Dmu143%MD`K%rM%0JvfhxEmFeHLfm8Duioj_Xns zJ%qL)-sF-g3J4tBvN7H%X|MSrj#&<|hDxJIKSq7&A5s4FiN#(^5JXtKKz!>!zD7T( zP~F<0SK@lk&36pXgw!Dg8E{@ai@XPhElBextMghXJQ-sByp z@)UxOJV!Gmr*5;E7$Uxz|22g%i;zECp0J&IyO2vX@!kXuKT9J{z4?_F%Q`}Lg}Qy1 z)cCHAE%>VSNH&(+z-9CxWsF-yZeWOLn3reBUEsxKys&I?QS9A~=x+*Rm_JUGQ0%_` z8sUT|(e^we(FHk-@I}H-nK0;uAVy8sgK*T!t&&~0qa5wHA}|x|EIoVa+lTe8j>JrW z-g6=Q9dgP-DAjEE=fj)jvZP8Rm95oQsFT`oV3@d*FzmDIibm(womaqy7fw&nJcNv` z%M|~Aw^pYq#A=gTy_Ui6kPk*`m^j?j@xqSd(jpXV8SOmGYGL+=1QNT&zYVpE#aS?k zWjXf4a;^^gpdc(8Cd%vVW;q-t;es90F}Zk_OMb8Bq6crT6qz=fLhlutcHIf=M8Nn= z9nH4G!#J^}yM!~B_Qu$c|6cL=h;`00>Kov7Xk?&dGpW>^`nW{74kJpupAovLID^f2V{*7p^D^>r4N zRg{rx=d5|SFXIZ;b3{|86dY`PB0Qha=TG-n;Bfz%uOU`t0@cZI)c0)laWte8{p6o} zi07a3ggq(QnSYX@+=5u;pT3)e4l#bB>QkCb5ds%IQ^>FNHHg<5GbD5@ykeVchb z7H{<&MKvV)+sY-+7C3`ypga8ET>P)UJ|YFxFpfNYl7Oq)I4qbaZFwB%+wiuQh{J=v z12)1r7xLczSGJ#;nIgx?$Qgm%8t2z3&wFhfNLHEiM#uuYN{K$0aQ24b` z;?nTv+FI}!+Y4r$X5S0ewr})y2gefU>Jss5xzPt79nww}5^DFUV%@cS_iJzNhvY97 z(m|MYLZR6Cb0*}BLLX!(Ri3DJ6$y)1VG^`pTF@|}zMGpBLiJ&le0^1*hgrfB^z+pk3H>&w%>|2MKrVV+P}+qt$Tza zthr2wo;?`??>;m~qom>2jFUfn#fRhmr^Yb>fe!DD3%u~vtJAOi18LNphdgoK)s9=^ zB`uFl7R7)IY$eDi0EAK1*1VCuanSVaS0IQlm2OUM@9j&nCEVN0kRJr`VvQ@qNhN=7sTB$rrd zK*Jh_#8y!TF+oH?wPrFY!lp2}kCCz1oH&u#`Yq8iPK?6~HN20*)pPiW(tq1{`u@u( zlIC3yP>(y=s^0NUixH$O{x~5!nr58%Kb#6drt`$7`A%oq(#H5}{O7Z{bVs=+l10|= zuRx4Mfkb2BVx3Wi`jb-pI-PA9C$pkQu(NpTL2kXEpz}Ols>fcI&e*>3O=cDo$WUfE z^U%+-sr@}Q!)DiJ@Uo~c&!q`6;^5tJ>i2%YEk@z`>mH$ViV!E+`mSA8 zpDLrNxA-M7puV=U9ykE=FwC%#kM)(yMvMBtB$iyZq%?u|K!tZOX&yC`bG^6UFJB zh#Gn`u&#v0+7%mCIs%PI;OGzjc}L^op?}j6Z)S5;|Vjv#5Ut1PwB>UT`_r0 zxLj_rV&PW1U)4dp6PjO{`kq?UKnkTQdCq;@r(OXY(Dg-aTaY{BE(UBa^cGu*g$tOH zM7kA*2YXZ}Fb^mf-u?rij_5%jx80Xcm6s3ggd>s1O&3A>cM&<1@*6>Qg0DBX=b9)f zS%-~Q$J9$nuhU~0Y_gF>YT*lU1~f45x1DYJiojW-1e)y#+*N6w+hF+GZvQXB|&nav_*;+?iX(erRl7x&AEvC-BhXbopDSdQX0?*%s3Xt%cP*vQU z;DNHMkcP<7Lgeh~xGyo5GuHAdUOsxPiBScogwy^|@1Q=?xdgb&e_YW2D1CQGo!4wC zNJV`HRYyOU(m<2Ny`ihND8Lu-+B4Ah`!C|RyX{Bybx)GjxF89E;l;#ze+#;}PtXsy zI{$S2Gd+ta{*t?IR=~2nSrQWR6v_6Em)5eLVKOAz2x7~+5&8EHGyS5Jc%zd zcFG>H-wjfU96r9c1LK%K2+0bjAJky8#HuYd%Ve&Ot^S&N%?RS6V zDSjuCmeyNE6KhTTQ<>-FZO^i%+NeJEqfi_{%MIsmob~xG<@N*g49%$u&2vL(<6GeO zs*o-|a@9D;kzZ5Ef>Z)u-4vDi$-Y)$ri%hkVXoZZY}GLDggj%g8%dWD@~IIj?0Wi+ zUGK}-OesNe47XX0{-G~iiJ830J2k36QiBagzY8s-EJ0R37 zXR0!&6HwlC{Kx4$X*p@}{Ec8{jE-gEf~qB6Ja$6RnYgcwO0j`1=_9{C87B7ASwuKx z+(Kk+TK&yERFsQ}XPbZ@>+K>6%=HxxA!@gR6t3GDNupl7+#LjNb0x@!8ct;ZWC%?1*RF^2KXG=Ow z#e&hW*+);K$N8?)k{Fd&jEf%;>%x^*u#DMJ*=V4S z)$2S?`j&HLGMI|veM!U9@&4bZG?x6fq-5Mhrh$s8Mk@*c={s>yG3}?x)s~X;(sVhl z2*DnY3o0Z=nhK$mVh14`8cRw5aON)7pSoa^ZG`XV9N=9-s0Xw1iH1brL`s$1^1!uj zQsCawf#a^$nl{*{XeIC7;?CE_7e+4NJmFgu#c#c*YjWcSdnr@;gfhR7Fh;p#WG6yl zH9(Pn8+@x5chv9LQdfm3(qsCj3Tk^$E5gI?QP#eLnDLJO<|ee5Y19vVYIpM9bXi}v z`t30a$jJ8FDGDM&AZ}=LT`T{aO0y4<~aSd#Vy+JJ1gNu++Jv&KEUHjb`;) z2SFX!wFRZla+HMO9w&_3hBE&O%ae3Gs56HlOky9xy7o(|EjX{^z*?4}^<79adBW(a z|ADP6pVj!th|Ev8wi){W`p^HMa{IHPJ+k#oQ#q4=!cUiqCkc&2DHhSIY;)dUA{6(% zGN#bjM4J!PV$l3mgsq}Ed1nzH+2tZO2{u`+5c=7M2n}jx40`{nC3k{R0k?e|-lFxk zA88|(4uhOFL}mGCplnCVS})NQ`TU|efo!i;g+0&1J^#{UU@2ohf&(elB0M)a*b!#i z)~ybrQE=nq`e!tpNT5b@?j%QNx!b#n72y^Lhl>Q>_tKL*gE;}W055!1X9*1Mp2Yt3 z!u|?UKlT2UU~@)wH1;M$uG>D}t2qHmpf?j;jF``5xPc#wcRAf2b3Tt>VbCFNEZQieCysqijjdiVtV3&RARm_$c8UG2 zv0s61?{i7{5^Ga`rrWxU0ry!fOEG=E2O91d_-Ha-6*qqj%p8$$~?Q!I|`SvQQ&!AFaJDdq#x%32IkK&33$hs6)1u4 zMfS@9dR&kT$!7t@F_+EY9^{od@Eq+)0BvL_27(S_vRS#}R2<-EY2@M83uHKUQTiIu zl(N`+mUZ^4Rh-!E3KnhN@81wVK*MMh(UE81$J~?BKmM9!NOWyy9hbFj2#MSZzkKvE zOu_=o?GR=8B$R+loCJ+T6S=SfW!-jKnVS zBGR^O8uBY+@%a90FaKZ*WDxABwGdpv#3}dKN|Jc>mzuYkjFDbye~4{x_Sc^4agh12 zJ5=a+C$sjekxhmBk_XdBPO;5jLVzKKk3r8-(e@pQh0GeC2^N@+>B%mLPsJ4Q$Ie$NgYbxAnAuD2N~PW+@Yi$IL*VpU z2Q~+M@_1n|R}j)k)AT3~11L^*`Fhyj+$i=6Hm7qyHyFfQAx|TRe-x)>5l#wq9IC>MCclLa56y9PM2Z4nxNo+h}gtlC8{0 z;~Xo7imh{7g5EwE0=D5tGX&d{Eb^TQN0xQk;oI*a3`0K?1A9_jvkXm zu(x; z5o=R{zu~wvuFZu%%Cy7B{ZdtkiZez{<7@k&7pTt3DNg`Ri%5z(-zLKH@xDy zSvq`|?&&-8Vm#5;o?>Xw=hI3ty|jk?^q$oawbSuRz>!e}%+3g*67;d~+{E5##OcZK z2ELK^&uP-UHw}&WL*~C9o1#j&{h+|qBPCvFXOr^$$R9}ajajdvX_b8i@Sht zhQP8z_J+y3OO2Smpjjob55r8C@_Gx+l6Z@g<#Q2-5`*-+(=}+E4vfvKqu&$Jx$lW$ zT9jDEdDtAf8?sU?eKb^hJE;45)i}`X-7D~AnvG9f<0 z0pKcN#KH`(pnt~{q+h!H&O=+&o7WDt&G?==cC7MG&uH=U>LZ($|3^yI1o*|Ii9I02 zpiJx-ad|IX?APT^UK(cI71LFy&H{&z zkds#ptUcH0U%ip0p}fs#+a^0YYTfn_su1yI*$y>of9G_LSg*A667kc?Zg*s*@$H#+ zIpiwf>THH&SxJ`E5jXUUaApZvcsylpE(8K1j6ej26z}iFDkmC$xdCj_FQ#y)yxhDX z@4ikicHp=na)4`F(wH+^GZgEuE&cHoW%l{}`cN&==z-p8`@jvA1T5D^n_lkm92qB# z{KZmoVK_!~bg}|n%~?ra20>=m?NCkDV)N3zpt0bH*lCFjkF)OIiENNuzC(ezzMX79cbsgvkeFW&*QvI!8{9>{K?@W>lM9X=f zwTNkLIr>tYAK@FjKlC9355`#nrU*^HN@ zL_vnK_h}(}{VZJ{UQ_D7>+fV_7Y$w1;=i?All;^1*AVnT$s(c~8%fohTJaxOM2#g3 zD!am>jztV|N@MIz58pW^l3@mXu&88?w3Jk<-Dp>Ka5wfBdXk1PLW(uS=DU#VuUT&MMS zVSP0IiQ2JtO=b+uWvWXd7=N#*+JXHs7A*L_i5Tm{6dAjiEX!&tD|sLDqNMQ=t1;X& zNg3&?MWB+6l6*Fm(I$$!WYM^x!-W^^~<&sZ`$!g7h63tb;Bm|rrfA{ z7tPOpj(^S03M8fr53FhP*8MU4Jr^;S#{Wk8F;u`$u%SpnaX7Aaas6}h}F!z;(OAHO$J2~`e1K^4~zjTxa zckO*!8aSGyH$?uxE0);8{ZGgwjYH( zBXrctm7^p(hUS`1mcj85yDr=orI-n#+ZWuX=leaA=7(t|L#3miGgzXWiW#!%qq8aU zf#$A-o%YJ6Y)1kr%6XW38-B3dSIqP(Ga(8dj?fUG&a$|az63;92#ex#DmChJ(j+-n zQ7X+-VIzbkHccl{XKB7J-(j-2Lv35GTmVN_1!-0Ml}!7??N})m^|e!(T^xo1d_vXx z1D|p(X1klH-Jm%jM_~q$`Z8Q^Y4j6A+nrCy*!F%rwy6_(+4yp70Ti#>RpMdy40H&y zqYu{lc&Tq#u>U5;u>DtVvNlx{+>*P1WE#48>{H>m|7OCMrnmU++ndaczss9$+!~+y zY60J$itra+y6x7>w{qX!<0}fsb`0Xo(7)?Cww;FW%ecL;VPH5+F0aqc`>Ht3@R!2v z9V$fdhhXY-R9gR8->VE!pILmfH!~%w4WIYrBCbzm_uKmP-It`k8zqx#+ffla*nu`I z&6W1jiRDHTmqO*P7a#=gHnCfT_%f6lPpa)gpA#V)ZLpy(vi@VPaxKxwzi``et;Ax`;EB$r238ODbHh%4WffyhvRk#aJV8|~YD$Pe)n;_Vdh{bY zRu7B2)ha5>kjr5~p4Y}B6Spqg@QcYd zolH&+%^`lBSVi8oCX_8$x<1iz9TFp&?}Zpj_ru`@4OBNjMv69hpZ2{8O85>!fH-W> z1%jid4N+YEgN+qh&0_9IH(5z`lEHxOMp&M&0SPU#IRP$+;S!sQVJInZHBMt+QJ~|2 z67C{*Fr@hl>)Q}y)YIy`7AV5ta5f0$m5b)Bp za)XE~dR|Y~duFvvLTrJc^=}&;LAov&TYiojA}Dx8y3k%$&!q1PuZD;h_wu7yGCpAo zsr-49fI`?yMcU$%Cchx>$#QBJwtuZ+NK#73P-(aYBsM5v!)c%NlF;miC|+vNc#>XL zkPSW{kTls+v3&irB5L|f+zOf9li=#T;(pS?*fX~~2M>BX1lj+5kusFrAe$_w*vfg*w`I*PR2TWU9S>8sD$K3FF%t~`8~}}j%^2?*0Y(~9kmNJQiGH1CA5K)Y z?hFsWy>u)g!G{AtpnM%P^$cnci|xEVYNE3pV3Cfl`7$e#!5u4hjS6wB7>GclGI~Rz z(hHFRnvh65eX?;+D#414?};eBSMiiU?o5rQn$3fv(CG&+wznZDONt6cg0vejX%*~n z+U`oM!;Sf}QNQ?S=`JfD0eZ=>#)q&vDp^e3Hzav&Zi7~s!eGA zA<%CJRhe#p%6F|Lc`35b9WLF9W;s*eE10}=xFy~=mV!#m^R`4SY>#1~9QL|=y3p5~ zO49!I&Qa{#+AqP4P1 z=DRG0&Fc5Cmp6xBmF&_<4{OH~k@wEpsF(4<`;MsJE$KS?Cl%bfCF^N2v0{|>nOhlT zX$Ys03+rk_C`Kou3sq-PPHVVFMH5c>nRa{;?#j%FL?hB(4C}09YNR?*#iA;n$ z0u;l%?Gkf0X++?IES20@km+2Bf>zTPjqsuI6btu)@j^2m$-svQ{nTAr@sdC(N)EK% zW@%z_xxliNpEZtuE@vw5lMk|G9?&&+$6pY+L!Z>6{sU3~T9bO2x)*{Dk^i&cCHeyC zu~au^<{pE1(FglDjAN1T56vFuOhThAg2`(@s`Z}~oEua#;}t<7$qqR1 zhK7n?0JV$|n{fc0k%}0(pWoXlK!ga>n}PxyNW=^tHCsQRj~E{pdg#)=ou9E$tjYcG z=XUZmgJ#49hjMfwlRCnrxi=uk+qtE3AJEF}d{tPax3W+c!DjJ|&o)IdNS9Eg>k zT+%_vi51(hoY|`^WK(#4&sy4cdA*jDptd0orQ zYlX4BKCw+C-dacJLPL!}#4Th_IMHbW-Zx&?(-{>6?gQwoGeJk_`1IE(KWSI}K&oW%F{I3DkZ6|fo1BX+bQ?)B1( zcU@8;syhrg1?y`qW48K2)K4zolxj%R#^f8kv{7u^p_e($ZdvG$&$b_&1Agz(r*2_FSzSW|3g_^P-nGo2O7d4t~O_y z+Luj7f3XkVyn~2y83PIDFF#l_-JWk8%2fB2gak;qK3JD@g+2TAA(t;Y?_(RM|2c>8 zoN05zBUMrUV73zhCJaKvm|V&TA`S!iB<{0->G!5oN1{?IQ{<`6HojY%&c;<+L5}Z- z8dJ?S1YBzwIDCKEmSTzw29xt&r)9M=B-*oRwf#pdNpOO)D^@8y| z$I1ROdbD(V6a7qa>hsLPnZWYGoL?*K@$>BbF|CNbtLPVb-KLI5ogZg*>DLwop|}Ri zSPM(Rd-9@%F{@N+?CUXR3=Uo`*aBfHx0!{Ynot{@&B4 zf^4ax7DmAFjxhsgT@f{%o{t!=^J`HD^ysZ;xso2*4|{Je*4$Ng%#t@1CwabP;;N{} z-w<`uPVvq-V*6aB39r%?K~Z`Xx;t;=1OIMBH)6n5KQ)I?`WuBlu8Onw(&AFJl*Z}=)BC~xO( zzCKi=iUepmpUzaF*O-usHuxu-T?NHz_DFtw^2v`?E#oDlN~xfA!R*vVR|4d2}5)e!W4u|+QOoX75=0tkgrA?jXNZXp(_G2*#wSZ{mD3 zb6XFosT2%@i0FDS$Z>LPZw;r`sCH5)F0?yY(K@zCrz= zOUD3NU+FG7VN8gZb#SY&qK?Ul{`j-OaS+W3+FgdZ|6OvUO%OSJ^V%BAqvW^=o8HHz?9uu{5F57$(}RCYz})aW*GUNVeSy|Cy^@v5L7OP%|68&YqG(C$^?vJOMyh zEFhNaHlbuo9GqjOILL~8h+JsW(72yHyCpL^v)(8vu5~~|wgR{-ynC7ZF&p}m`ewJI zcE9!dltUJh*@gaiIH28ssh|5s3%#S~Q~Mv87Eket&E_qiA-O;SI$uXgc-Bx(lT2J3 zo^o6%po9XlAW;0%4Uy@+56DzSvbE^g7U+hbiQCv;;ErS%j#BH%DAL>0kadqA_*vUj zUGP}joV&!^sY|0W$hrkOv8wI-1#T8upVitRX#u5zyRJhk^vVTfMe?{jOp%Nf_(Dz8 z9bTyMW!sh#3%Qll(Jc@HatJGGeD7M#cXprlI)YF6Lym$7mOxx&#_s^5zyvt{e!WXq zRy1kffcRYIF2XtF`2o6 zCy@{NW|GEWEh$c??k&ywQR!51C%Y>P2U%@e_F7=bAAg~BwpY_^z5GweEl&Un*G5GD zo#L79hOQBw^`2d^u7S#0w+@IZLCo{>T5STLP)3kF03NP88i?rf&T7mXUnoo6!b)_i zwcQ-`##?R%qgM9JN~o8Sz_c%0^Mf#xQX4^Rs*oulMGP=G_$+V6-MM_7YeKDSid=8` zXtG3wN*&8dOZ^q9!ZOU8kOjbG1%x09JArJHue7Y2!|r4}2{J`-W;l|Nck(}5(`ip{ z+4&%-rK`q8y#Gz3Pum$#9m_~OOeveJxnYL{{&DE;I--qgaE7)wM-3~u=k**E6nqlsi^Shgm%N~?p}P-*Ym|=hykYQ^dYUJZWYn=$}w@rK<$z2w5=gmXlaA<4d0_ z`=+PBK7V_&N2ch39+bc`Kg+;xO0k+sadEx0K*XZ>r74!2;Tj4y2QRb(ir#ysN`@^W-loXg=mS(M`Sz1` z^>s>EO_AAaIyr42> zd*Irk5TYjfPL*yE0$(jk2K3+XS_JPANMeLBLa9SxT5;2Jocbr}n=0 zm#*gt5y)mmy)A3lkgw*GlyIDU%Ko(Nhc{imGi|{SE1nPAGvoN+S@$$>i5>}f=9@5c$!D9LpuO|tAw14); z6)Wb=1{>3;a_1=dFX$*-@EQ%;tua6FE83$q=~&9?$f5u_GhMGIwN_HL5T*H3pqT zQ{B<>cqTCZTLS>^F^%N)JRlsxEMHwU2&HixYc|{?1OrdF+yjtj&K+CKl4m?#LCMLj2i5K}{uEwEfaBzG4ZEd$ysHzBRbIeRbiW0d;W7+d^Dea=O5e=3VF!M?P3d_`q?A`T?6NKueG$=1|&Z2zOpO z7uqozdke9_*UXjzhaKNoEp~p`)q!he1m@^YX!H8uU6~}pJVe@CHj}irpB~Nq4=ky}Q=j$8k8{P!3@2yIM9+ktXz27UZ!a7p*bby{}$o%F}xPSW~ z&2+ftFY78fNNb~B&i}FiN=Tv2>~vlsJXrQvVl|Do66(+iz?)^pwP>!VVwLv5Yj-Ve znr=gK-TFz~)(p*(P5h1O1Ms(W(&C4|`l89Pye(O%KG}_l;KRvC(y=M2+c#MI%KeCn z^ff7r%6WjHNCeEf^(qPKaI(N#UnKOTssM~3w%IAOr2JqdG}zdccwU7Wi~SxW20ALX z)X(=(D*<7gZ41)5KSdmv>^#>-NN}7tuig)!w6bVY2za(3j%==Xrt%xnNe)Rl`maz( zHynPD<#E4Ff}huYr8bsZroqXa6(d~4o$-mx9=awA(k<}+%zH(tS{3to^4_G4-};FY z97Fs(H--M0{Px3UjkG0Q1k8#PNzwFyPsCp4zw9Z%%3zl+r{MX-a?x8D(6%7&Ior!f zPc3_#{zcsu`(205FPpI8Kv2GwwBR|hY=58fucm!@;~ z5sy?MDT%X-g>)6NT_Vz<74-@I<>!UKlatWQv0QHkMP6jZPo$Y3l_~1>pwkl0$nf!R zq?Y3z7CjwvK&6T3UnI6-_@izltA_L#>N7nTGxFU-&dl?N=^$+CsxMM7TKW+Pya`f8 zYP4!oohWP%rQF%-{$YxCkh`D-Xp+SMbl^knUCb(Tjf7@!uoCh|Bawvrl~wNwh`megP^beJ zMi4=}erKgZ_ISe}M3z6LZ2AnhDzv?QN)t?rZa}w4xc-A$Lpm>R{VtwBwGrnsZ7y{* zHYWqhsTd?L6dMnlA!n_*I(xnKpQspcO=B_5H%)KG%@F4qu0?g`PD;qhB3+9qJ%fZc zGpgC!2rRU{1aSP;C_i&YN$L}Q$I@vw?OTBAlr|?A(`eG zx(h(u7^jka=>yr~_Gj*oF5m1PG8)pH+jeHC-w6FvU?ZCLdf&UVyfJVa;tQSfhfW2r z-@aN!d{DMOAgzFt{EZ1Gg0_ZThn2x7u9z>8l+~}z#%g}W@XfvF;^+s?VzY6!$ zP74&FGSK2kT_Cu3$&~(N(M6+;@q+LW)&0*3`iY&8a7}znu-wHu=@75#ruPW#LYl5W zm(r(F4KO#-Q%m2+P_gDj8BOprg z2aZdN5(a}^c(09^)Fw874&y=Yeg_Yq1mqT{$ZDww5L~f;y;UMX7p;`C_jK#1rspiv z&~FKA{?6w6tl&2IZB^&Ch9%1Xg%y90Gu{y@#PI&(l^IhR30!+^IPjagsd8Kh}u)=PfbH-gal7MLnt5B64OkpV!gD&4toO zcADPv%MU_lXygTFS&&xBT74R`%`qIZq}NBE_bSxLHW>Va)VVNDW8^(O)0t;YJ^gC( z>Xzc5BL?qwR6(Pp<)9+!pyV?5{%&5cQK7-%zB(W1(6kokfREN}V&@yT`rk~q16sT=yI-+qk?S3Qf2#*3ygJOROpF&ye&9HdAcjXqJ_?G`Ak( zwzc`l@n*4t_&Sbk;+$SWixL!qRJg2H z=(F&9!s!q(yS!#P%Im>g``hb{qjM8g3F=HJ0bab{6HMhDK3&!1q zLRDI^?y7MiIc9Vv)2Wk1yg8^*4$-F%p!IHw+gFAcv&TO{bM)S`LQ{`eSifuX>uUOY zKlxR`k5m%j;4-w-Rh+jxado0dZeP%XYI0WTe~NIpSLrcf4Vv`d?_MQ`O#iZ;*Ss0! z938N+ZAM5nSm!=%PP{unzxC7i^7x2Rco`HEcE>JBk9U$FBG5}E&G-Q1J!P03Vs(5Z zthl$qJmgh8eqC==SJBmBIC)ES<0fa;qUiaoNDK0W-vD^tSBdn{w~UUrT2H)~q0RKJ zJ2N(x`uD)k$#wa2*UX2mC-1ss`Wid(2tT3sH!lT=D2-Legl|A%f`jZ>Kh5A%y{7QQ zdsD`f7pS%Vb?bQEX}Pc{rVJpot7zVDp)$sE6!R3@hx=0MW4*7!pe75|55I=CvoD$@O$dq<>>Y3=UJ^|)mrOH_S6QwgYs030fO;Lv(4MBy-s{vd&?f?{M32+6H$E^T~ zt#FekKDamA2RYiCE_J4dbcz$mY=KMke*KI_Jb~jd)n08?BPG&p*c9bnl=se#eSR(s z>!U?cH>$oDY*I7M%YgRFTY+#{{Q=Y^rAv~MEP)oi!lzV;7k3(85N^Hrve2Jv( zU>qhj0&_$)tOBJFXv&~ZVUVpWv(Hf=E0ekAoCu3nY-l?7tPNaz58P<`94=(%QWN=>KRZbe5<%v+|s zvKGvW0CBp&MF?!r`0s)#A|iXW`ro!X)~3Ly1aD${w_ri#oQZ*hE8qDKpF|*k2fZ4W zE036F2DAn1(_xnz1TX~)Q&-lZtePV1V1^mF@ML+9IhS*P!>wmMi4pBuz1wg0$sO;7 zK0&4Ti3JtXuXm4hp{|TSh}Iv-!q*f0h!l-VD)hn^8kH+hA1~OKV)NW>IV&7{ZHuK< z9cgmIq`u`}B%B4AYS(7U zsgXc!Cl9Tcf51jBt3L|R)Z>Ne`x5f|g`LWcD7WIAudNwfO9@X6D%7@5hWfdYfq!~L z-G^X8kjMP>|JE13-NgQvw*mAHtS*W~AL1jn97ikAmFYsK>wRZDQcw?evNo0c6@C^! z5d|gy{%`WPJ^+ELz8K&F-X9{;Y;iuvC8Qo#CbyPSRbh83V+4$q-!R(X9g9R33wkei z_WDe@0Lglk1nr63PF}gD(K*XZ)!EQBia2O1+OcsfU92=^0ilGjbg#76B!ju z#*pr$!YVA95?f&YR6C%rQ!&C=V}r=kGE~r&#@axwN8vl6b(Q25?1(8`32!HjWetW2-$<8ktXqlw>N-Q+Dfn#;t_-h)oN;Rbgzon>qX~B6^Rs z>ttlntJx{%(aA_nmiY%&SZ=4sFEN9BzBL1O6D%FM!BF@Xwc3afKk~YFxq}IZqk=S5 z9h`ic7&S#s^^v5>o4LCC6FbdEczSr0-d(!eMtC_klmPV_gMbQ^n-a%BFIVfn>##h( z&Ly0Yu=p@r`w%O8^R5fNp_&$32EV;Zd;K;QFw*i2)X1@Vt}LCvd`|v9rrt6t%I|w0 zhG7(tlK9X9k|Nz5f^>t@4Fb~Lsf37h*8rllbPofAba%thHN?;z&*1m>to2`O7B6@) z+?@NIy|3%q*WO2S+k?CbhLoT0Vv&`-yGJrOMs-NAf#8J3pAf2T%lvkGLt5Cg00rM_ z4n{B^18CJKvoTrt{GaIL21m>Iz_ZJ6{%u2=N>s#DV`HYHDd~FMs6gIk=^PyV{p+7I zqdW%x++xf@xaADI3AquFmJsQcAbZ8bcdi`V3MGdP($^H6ac$4q$3v-Joq~aONdEyc zy?lWhvQzT%uyQB8wU_aZ`px!-f(likPd(;TTme5W##QF;a6J1H0pl9 zqx-GV_`#NBsuL#U{6%H_w1@u$q(N>yc^6f|{$6jX4~%=T7~->f+#TAnax|;$bCpD8 z!w&Wf|40%FAK&LdPJ`Qu1J|#5UdJ9vkIMQ2b5HQk^WOglwOX{YtCw zT2nhzG3uq_8^4jqQ}^b_WyrSw$o-O-#K?dP!l(5%0D)L^YhB*H6%C3c!AevnlpcG& zouOGo7jnj!8&%Z+LEUL-ROlc0d%(mUTl)NUQZN3|wh(igw(lZ}dlS2`vGw9j-s-Gl zz*3fP{w$rc-ENzN?Wl$cHYml3f5vR4q+KgY>0ll@jNZR>fje=xeGZE&3dgkV%su^1 zHFk>wandx3si*pb*!w2ZDXkWlX=C+3A0Bqn;I)(HWCY^Dk4p$F47596TlAj$1! zi+{;iMpid!IALIJIQ9NV8NoZ^^3<2XPEip~Ttt}KjQmyD0&Z4#^}mvrml?rG0wt*5 zni@|82@?9aMaNBu^Qln7g4iJr(bfB7@p8Hnn}&xs#cZ`Khb>z6&)kH>$Clsy)C`9g zRJMz-+f((@5q)y;i+se_Js3rF2s#I5IksD(BrRd?supdK@V>z$24xCG%i=-SR4GOO z*N1KXtgtVWg9a>EV>u|cyP)(PoG;xPjx61}LN%XKMCbyx-_3cvPC0?!jMA?oQXBu0 zq-&bzPlS~XgrkzXa=z5S@`0?j6_xoNvV=!%f)}$gEG_VSO*viiVu12Au0W7CHR*M7 zQ>%PaIRDIH`Bb*x4b(B!o6LOfw%v_DUKY@?h2~Ho>=(F{4lUDG730ez9Od-<1O_jU zgd4l9B^i#(O3`G21n4`GDT>ySr`ox{~rON|Z{I0&_6 zS=8r;=u(TKcX!XN4PNAjATe4rqCGW|v1enYvS}o;Slg)LiGF`4k6l*)A+`hBrt@#|^DpOZ9j{&M8o?N0jMwWzst&cskG&-6G_ei@b5%gLqtHiaV}#JJ9YeN_F! zE!weaMPVRn_?8wX~f`@*wt&*@Edj-{Lx27h4mn+>| z*_-m{5Fm+~VNBmXXKDahlbx5ge`9(Z*>;!yyUFL%gsZ_VO7k4MCV~7L+4GZI^Nt^g z^l~1UU_W>8+l-kL$={$G(UXU_*SOKgS$Y+IH>l5CZ{Z22%HsCD)J~L}vzT*&mknjR zF@$u(D)9&3j8D(zB>KvC>7BRXm2N6{BOb40aDn%a^ zh8a}EKD-ir%MSlE^nnR|YU#EflHe54NSZ2rTmCZuwlv~f$9tCpB!~*$_3^o|m|j*f zKTm>0lAtd2evlzHUwrRX*yeBTzNND`kFDc$1M{2lasL)nK>+i)7N5iHjpglpvB22; zt(=O$9kri5%r$B((U6IQ-CAwc=MC@ye|mslFVEv-8&F<`k(EIG#^~i}*EFl1R54$` zKvCguG}l;Pd0dFelpDnBe9?qO4W}^4=&<7a_Rq5Qx#78NZx*iAF#7h6?T> z1L6OS@m4^C?kJo@P-_ZW&_%CQVLZGg_WC*dv$E6Da5R%YUMU(XoZ++SWW{m|>*K;c z#^sX>qA{jDYsN}vnv*XTgJOB_@>@%oCgdO4c3RqZ#RkLO;xE|6{afEBbJ=g5zGP(& z_pv>4j~xt}mDv^~c0ieIRL6mc{WRk){SLV*FrtdK|MAkI6CrGs#dHZ~`E~1I!5=E& z8MVFHR(9lW`-}g0VMpNAARogG#0-o6B^4^Ucvw&D@5Hp=h0ql7_fC6+=^?u+*XX5| zuk9V3QySso)qzu^Huqt+kzP`-!$*96D5&FTID63e37lnwS;BZ$3{?Fp=ZRiB- zf0-7T$W?=S7Ui%g(pnp zxVE-f107Lq{<~@Qo?vgK4Cd1kieLygN2MI`@2OE{%07n6uI6;fe8P?7js=!qzHMwq zy)UAE=T*7mzQ{0a+>`q1 zBW2;GENJAKY{q*S7Z3N$-6-TQ#1c+slO+*1s}8u6ZG1t`{Y?fa7dQVKglKn2pHG+1 zfVa4FRxrOl9KOITJ5OmEJQYCA-bea3IOxU#H?riM@yVpX)YnHpNcBuarK&k@kCR*RAiZm#1k4 z+FTJCwU&v6VE!<9F3Jg8YzK4HisBUKslTcA#X9N*Z-sFMo!7Mt&f}96RmB#wBPT+8 zSce>Pev7cQQaM~IR5#qVy@JAWd}h{ejicu#i=vSmNfIzd>l~}BsGQP=nb^4(GQ!)Q zE8^wep8d4RL_^lHKP5;pGVWK`vPoymrqjclRB3lR7wxckS~${gN~q*aAPe(jpA`#Z zfSTFCyHw_`JS;!tOJcLvPt_zzsO+NS;xO35^MxypGYlmXHQZM!TI!xW2B*CiiJK#2+g6i5Th**7}{34ew5)#@?GT4#jLk~ zyX{0nAuW-D^Rt*{LLY!LtPmDYKTSHhet?s|cuO=~V1KFKq1LzWC7edwiome>*5Yj0 zF1!co{SYCWI~pBfOG@#>=1niQX1?0DaI|EjF>r=lF5R zN%CWyOFTCmnXT6~Vue+UW(Q^r5(=q15GbIsO(CrNkC2a`@vb1BrA6*)jU(d`_g80s zLsIZ#3!k08Y|P7iU(LKWMt@rhL&^BlKACn|W5SBYxD;Hb%6vA<5aCO{%y1eZvhfj; zh=?0mjU&xv$k?YaCF5h#Z=0niz(_{H$&q5{ab{4Y#pYl(#Wyv(6MlM*E!jV(xDQ_%K>iGwKnuF#mIq^k!x(PJQ!xI)2#1xHokp~yK|_|WDh43nBZK0EaUde9~r_^sTSR^P7#PZBk3eP)=Ss< zZTYfyUpOV6=XklxjnHXr0<$eb*r`pSC3>gka2hh*{4QN;b%x@AT=u&WoRB(ny1#Ov z3kxxomt2?H&tk_$zO_gnfgWL%G`G#8M}6J-fsPZ-z9%%51)t0~+~%A4Ss0=;AaW|F z$>SO6U6o+%B#=%60+G)+E2&UJ9CN3d48TcmX^ewAz27XJOW`n|zzTa-!Z#60jumT~G>o zM&603^lMZCnh+ZT?-R7hFi;iO?AV7g62+U5cuu;B0I-Xvf?nB)Meb&1rZqA^PywO@ z8JK{Y#V!w)h)kZmmvOiILa(I+k5h5dV_ z+hmw=!9f5=vjm^m)f|Hst)Q0?INRr!%UIbv6k~C(Tkf>XZH4h?gWmIRyF15RRlxGp zeu&uRGp1uuiZgbLn_uODPWOLa)>HS%?)+qSb!zhdIyXAU<0ry=?io zz3sJ)az2|0k?!(P5`jdvua@SBeKMitW$Wd07jo)T! z8fCA?>g4d5$HP)fxDu&+sf%*LhCQa@_hg6dkKtqs3?Z{mY2Gw+ie}!gcNL0F)@HSNQ5MeO_1{AmQ0V8VI(ZekKA^9P8k1YfY4^n6j`g)lW?Vmf+nT%!5{1+i-7$8~H6XWpEeKqpR(tia2GHD~cVgpU*<`tL6RG$E_YS#v1J24w&0}?)m9+HoV7)8&Qu|Dov?w(I|?M z0z@i8H;T72-kJ+uc^`>sBkT)3yf_N?6txT*3o|vzqg}#@AjkFt(uK@Hjl%aLK_9UH zzzo-nwH&Yg9_z5wQ%+%zQnq*ex%i)y3CiWluI(7Gmy`-J*zh84Q3|`c^Au~mf*V?0 zQXF;ANxh@AahT5!!zP zg3rrYYA&t&a4EBamnktP_v6l`M94$|MG|g1F|Svn;J}r0PU|zw9l*eZRH$|InJVddMWS-NXeP{o1n;5mxAP*r(e=kO^?D+Q5Ydcf37>Q#9!hRLmX@K2npx_Joh8NcdA@&AV^7=)3_-mJs=!F_p z%oUlcV$f;7x!d91{&bb;<>}sgjWBaZhC~Guec{_UO71rj)!tu8jVb{3{oQ$O7mA_J zj)Q-PxBE)`0ui0p^eRt^X!qSsTZv{DzGb*f|MWsHPiP=Aeq~qY&n>9H+}?^{NwO^u z%p4|6Ag}hcHAY6V^Dm?&nS4An5p^9{t=yp=vSpK| z%&FD-&6#SRML-&L53Ct^sGh?IfXck@B}B5*`1)l4?DDygCqgGGV2TN_V#BqI3p zn)(DNscn1Jd_z?M8?azmU7S~@7FY)h3aTFnFmp*a{27OWDTv8Bk7CF^5(m?idN8i< zU&dj8V6-%m+Hmy=SuGWfo%)zTq$U)=EI%E%u7%a(le;dL%(nDNi(*kB4> zM+S0yK(19&C&gxS+n=U)=Z)+5**fY@ZP%{8ykkC7DsjTp0`{$Q6p03<@VVXJvhC+1 z2cX4?aB_*=77L^F%-_i8h(lK3claufYKAj(ysAZ0gtP`Twzy3Mruj{*N1AFXYLk1) zhwc{L|M*|Iu1tZsIVfy!aV|Z!NsndEA&&%nojbb&(~;Jk`)xC0D(4NK`4@+(ZwwZ# zN}3)co`;cytBT!Z&rqvd2!Iz~%l1%<+RQMn`*`lwF3Qu~bs3!p7QOz&cUMNmr1srf zuj+Sy%G8ZtZz3IMemA%$1A=(FM4Sq!Sk6bhK>{$v;Bd7&s<>b*RF7p5t4Q7ba^QEx z9V`#LVn;c#@fAv!_DqUVPM_pWfy#ha$L{VK=lPp`grze*%OoVzCg}!R(q(GYe1$Ta zdm{;p;OxGkG9yD+vwUmNzS!l`eVcpLqNQB5*t85Fo;DY+`3AbjX1iGd{;wjXPf_s% zzwy8$((duAUt1djBa21E;U`-@N#FXsr8NuqMz76cn8@}YEl?(m$Z#?zTmiug2qwES zpuA06wzC#NKf5oRNoA36)0v6Wg3tNXX8DXq2}Q!|V(OT+4s*w_N2XfPbw{^ba0)yT zg%i?KJ~;Xt4(0)+8tpAC$u*SQmF(MiQCkQXQodZz^M54@wEX?TVskbW&}7MNM0`i^ z(Kx=Ty~i?U)WF@XfLx;Zo%$~s>2xE((fz8wliKm3DVLXWmd9dAajdqJ$&ZgYQN_=c z3J=O`oO~3VN4k)~4V3^R8VYCQ&au8?!{8+<4-aRhm(@(~E1*i5NgX^%AIIv?S~orx zTxUb@p^<+)oesS)P8rG_!Mbeu6xML26Gbk@=GSc!bEfQPE#&+txIKFdvk&w2C;%25 zlT2BSO|11*`Zp#{Vr4|3tiOpvpa;mId4tqGs;0O};4}I)1 zKlGuEcfF;6MzDjiTN5%XSTxtRSO|XU%Fz%}F+KAGGw$6@mciTeK2_5%tn#q!8k#1x@H6513s^J(cUyAN zPY`g@U~ZAGRm?sx61Z=-Uy60PPnYfoR@qtsB&Q;hmWi7JLkVe}_X)VNGIkc$kVOx) zQk`!SY9+ASfIZj8!@drkf3G$yc-L5OPe@eIjc$Ttu~(dbbVP-M=X zoTc67>VGUd}!vz=#N5h_0U|i;l$6GAS%eSOQoK3W$O3_PPX1{WV7P z&W!zr0L>%3Gt{NxX$FNUKyD`IGePGi7HCV|a7#8E(7g9G!4T*Uxai_oy7jLUB@LWb zWR$H2Z(|B{Eu(2R9wv4iPz)eR_)GOzJot4xJ2}7QG730lyNZ$X9vzeCzS*yojRQ@V zmuS82jj`R5fi)68QirrxM>S;cZ=Dn4Y(9)h!^m6JJOJ6B%t{gaKQv-mDZJ&&3wLRo zEn4mIOlRI*;#Va4#Ju22+jd5vYWWjf|~n|6qJB_ z$#NWjmhFKsgVn>`9#|rnBWZ`;OOYtK<$?(0htU`y-V+{5qQRI$s`}Z64@`-l-$FUJ zKMx(@voxS8UQp3oztmo0T%wKGzmTWVyA@FX9qSh%*nk)gXWmsn2ISrACbvoE_cUPL z+YGA1^X==x;oxxkZv~BPK;H(kQfJlqg3L^Hm4$JGmI|U#FCqeYLqPNTwz&_f9wN=M zPBf)>n_JNppHnTv`bVCNN0VnCdy$Q}c9WR;>FS$xw-phBBe@l~^j{x3K7eh~);AXw z9D4{R+&rp+dNQH-TvYshcGc-|;HPVf z%MV%7Ia>Z!>Ol0nm+v3z>f~eEA(wqQYZv!IycoUzt%bAkDYGwh^vOnprmlo2nWxES zU7&^?b=D{~)I5a@5&Wg@H8GT_1NcL8*P}K?c}=jO^o?SD0_F`|trs5jaji(zG^sU5 zOb!G9PR>89z1|Ax?1%hXP{}CNfh!t_sdXvd`9Oaz^ymh>uxBy^usp_L{x#crMGoiJ zer#n%Hi?M&Op#&w*9}8P7*5a(?)PEN9AvuvUkMp_BSW864tTx#Byd#h$u^=84BQ7A zo3M8CB<}0mX=*ZEyg7b-bnfetEwKnWaJfXgG}!;5VHPKg zG5o&=%L?!5K91+4Nt3juFQhBzSV3wXgvF0hpBjty4SCB0cb1=7;ArHi;OTJ;ysgmN z{7BMrFkr2(DMqqeWlwl?s*OpZWla1w3WZ5k_l7W`AmZ*;AgASy1$0vz=%9v&EEj9x z8r5pkIUAjcw(&YE!%QUlwYb;MBTGB!s=V8=wQ;`}MnqYDO<`t@g_PTP2dE=Hk`wO; z9Q&@sm(QR=;0D%nmRd#JJeUHvX+`RCNOhP!_mhD#dT|%baep{T%IQ2DJ2F-_{E4m!}yRQbp8MG)^Cf3v3=rdikm*i_Srdg48urOHJhrv+PFtZ#K5O)3)23 zi>{6*l_v6hd|@)3U*tX_#thP7x@y+_0!%NBhF$QT>JG$(_dTNHJaCb&kQf-a3n-dl zN*)LYLRREB_MAw*nx)Zhe-+Jmo2)eB=Z+t-ZX{jh-8|9qv zK^;!>)CsmCowEqp1JU_5+&qq@>K>@Iq9l2a8Tq!1 z-So}v)Z!1PQr7K-EV&p%x`v{*nX4mjDqyP-1ga}}HFoqb37%?Sq0wRTG+Y>0;3n9F zgav?x9P$y5Q7Ffy;u;3m-eJ6?98=ik#32=&G4v7?@Yy0SR^bDXTd5jdJ0?a#vbiIQ z1un5>(3C6z$XlpcJLo+w!6sY0=O4XByp8#$&tG%tj<1g_jWKi}T-~o_OD~)R48ZNm zuL(sR0v9ze(zDiuqV`L>(R9$>I5o+8hWW2=>mO#PA{1z{F?n-n%G*6u1{j4EuFt0U z8PqJKpaM4Dh}l{8~EZZj&9ub_Bz-&YHFQ^ffqpDw2>F-IzvN| zC#EU0@51}T^$k@<`76vpqxIrGUkCbA0tDaX)HB}#idqipt>q)3dAqof3%*K!8Ty5} z$5)>=_^NO~q88(bUJ4~qYqE;yPxl^Di9g(0krcWTWArF;8Au?n3E1oGZ%)uXYgXVS zBy;q*n%G;*VFxrTG_vz9!H1M{C@UMps!<+sf#4KeDS_aC-> zp?VDKV-)oYfV`*5!+9}>F>M7FL^Xvz#FBB;l=tY^_7Y!|L3XsGlQbhi9~eGDf_7Ql z9<$p>ge=VL7u-DNHFw69vY?*oqfe)rG@x0FBP1%=hqWi_I^HOC8Lv5*M@;obS6f~8 z(Y4}~xTKJOp4*(~DciOKgZ}r*6aYxraWjG{uKSk|Ee$LqE)yZD*FE^Jzj>R4)UyxP871dl@ zlevzzsr1%bHuah;WE^zIG<3i+V+Qr{Z-B}pA4-Lt{bsJP)y%2bKxPv@eR#X+`B(bv- z1^}UIpBXC1C0tC8dr!?;-qU&iFb3}9OTWRfXm+4J8VNl9!8GDeC1e7+SB(54uh9ep zu~U3^Csf;9gC_qcCUlii1XOMdLx`d*8qi%)c|GV(Us{1v{Z#oo&r+hfl6DB4PbJ&l z@pJcB#4z8`S^Yn@0ReEOZxX2C~K%gsJ~NI8VL{67O_-kFL+!Jq?k)u^!P z=&1Ukz8GD9K;&{aR?dLqIa%bgS_2ikzrLq)l9$qNaXn-J*4`ymO?;_h5_zmePi^Y{ zC4a+>p!+zNgMn{eeI4gXvlBt}^Ttsdmhc7q{`_*S9Y0YAN6GK|s&P#W>Di!K9oFS#;5OX&PT@W!g>_P$b5}- zAEHUwufZ@TqJU|=(|YkSyslOIWO4%Jq^UnJG}(ZIHVL^16!>v$f-le=E!mey|DJ@? z8TY7$l$azLhefAP>IhYDar%tzq{ZWd@p+R{$MjAqjKZdfbflF1`(D$1x^iSOetf~6 zi)PR_I+`1bzz|mRy)AhhJ@TTw1V>^&r{7Iz>xhOQ%|pef=STZX3o;)L~zPX%xK zmj-}>WvQi|5(?w5@wgKk#@6M2nAv{UW&g!RU$z{^(WZ^xpsoCV*7~0|>+`$mDV-i3 zWI~NX)?}|#XhP`eEA?8eQV+}yz3)q)KO%;ZSqN;m2RUc141_;~EC$x9R;${Cl2MAXPwfT5<(Ms{$)C;<-z8j{;+J#vxCa-#Du^ zh2QPK^I${OfHFQ`iFvEHoF^=910*S^;=c08>w+%rdEa-i@RAei4m}V9rfQ^ct^@rx z_7kA7&n-yRgiW*}pd$t$HC+SQ4Lj$mkkB1~a2RkrIoMZLaNTD>=?leIky2Cyznju; z5D>tW>xs0azKe`c>d}Qaz7doC(DT4kGJcE8*ElW>@$`RWQh!I>G5Myna*%09Hp5hU zYe!aY>BuI5N&sbPQ=XTL?ZTFtrT2=jRs>m8K$3^xf5w3Z=gwD>d-@@-0SfPP7ns%*Pk7uS7ABvE ze~VY+?Q$pF{EWC7y;Avhi}l#hlHI8wJ~W}!TEG*MZC>j&VgOxRMm6$pDFdh*Gj>ci z{`#hjOHA}NzhzB`Ep@+skUfI^=+5p#s?j3;0@P=buoQLxISd@-4e}BFZ%;PTe!L*O zEAN+h2%225ot*x%Ov4x`=8iQlb~gv7?zW&Z}6B2SZ!1Ov?J;%pbmzz4b` zsP|=N+*EdO$?u;Lie4GJpZ%}f z{!ImdC&;yacjQ}7y2uGRL_oWu%=S%TxSm}-+WmH8!$hMnXI`YAZ!WX1Q+Zyha1i#s zIFSuAOOAwtUyh_wo8)o2v=GX-Hr}fL;)x`n zSWYr$^-OQrpro$=@n5yomRjK4^Q`kPZIs&ce+xf3micd8r_og-Cmo|HifcAQ??x=JVx5$%L1qGA9+5Tz}~wh&h$7zfN6a)m$PgeGtAcf4I1>?+=c~iGL0; zaOaz{A=hzN9tp1OTznE)({2;xn+O7J2#&hVt(Bq{P*SrFC$*#jyQ-$)-iqM473PEP znh-TA1ux+$>bCGWzqJ5s@Tmb>-CLpa_;H)7G>#xN2uAbdGzZYirG_&715jB@ib#X2#dfe>n)OAhm>ZF(sylMH(1U5o28} zXF5tEc^m8B<8dkkJxgg72E|J@6CjNZo}B}A0z4WVpW$ps%MtP_RGfyXngRs^AD%pN zvEt)Ig)1+cpHi>>wNkeMpuPn9EqyyH`GD!=s{$p zmd-9!AdQrj?-@mzi&b)mKA6t>FLQ~zHhB|+2cGhlyfvVq6<7K8wHpAzk9Js=;v>FA z6Y`$MxlS~;B)-AON*u80TyK;qxUJWT7M6K*JspjG{*)wgsQVcE{TX%m&0Oy)S7> zTb@X>JWV`i*{nSo{qA3IW>MzDXFR>AD@Pyhi9IyOdj2iLi02A|T6!-Fkzvg8rT6C$ zW!67hPV@i&ECByFmqVY+U{Z02N(U2OOd14~WT&g0YNtCWj2{vnctCr3RtPj1!{vg# zofbNEZN^e3S)6L@Yn7~^(x1{Ql^9bGu$~18O`6&wP-wu)X_}g;lKBE5c&(?eOb|Cv zx0!3Ww=7^+SyCD{MNEf}gO!nPs72SBT&$ZBo37)a8WNnCAb0S0j-Yx~z&k=xZanWT zX}2ofP-fek7}k>j3&$X96FX~&5TCEU(T_?^OO_Stjn$n3TgBb2}oAl({w zyBcf0>`1Ra$P6mf_5!?(Fg>faL37UZCev{G;P^8!v7*=N8;_E%P|U!ZKkT9!AG()8 zEw3wyldsbhTa!%v3$dN#3NKc81>i$R8{of=*=Ex_bEJ)0eiA9M0hwNX-4#@1+v$)A z9Bg=erGEOLD|a@~T+Q_4gDkWv4E6W4e&z8SPE|`fEV5CSQ z^duY7hczSpEB2f$3Su@w?cY={6AH=6KLL)X&oAPTPV(({+YqGrad;sbdJbCi3kLUBa{z4ccLOJFavEIZl8w_01 z$H9Q0xLZ)X>m#xe>r)rYcW<@)&RD3#ZOFT|dxeVhsuZ3s3oT?fRPnPv`O2>)V<%o1 z+$21sZJ71kRY!UP9^@MHD4d(mdg$|8pP%`anIWu10fzL;OM{(5>l6R1xu3K%cKZ}f zGwmyxnJFChtE(cs>*EppPkt4J+*HyZ?gFGg1l{yA)Ha0u*XNKXvqVXMjK#Z*MY_A4 z1M?4u#Zu5!xvz z2%kl}|65|@*%JuS<8BZ|-a>C;xFUj^O)<-G%0Qxsv8DH?%!g>1^2v`%6S6^%te0)v zEza)oGOBA}=S)*z=Nwj`{_vx@F(ocXFDG?$w2maDHS%v3VjG{IrbgkIbDR}*$YKN^ z=(txBmxv0t{>$#%1f%Ga3jof*`B5X$03uAs(3{}WRjq^YkEy*L{#Dc+<16}RNCXiG zKvVz@Dc31`0$dS%;?FOQhf|6bNC0j#C^IpsnQ(%8M<4l?1NM71(s6yb9{a5}>rxJzF;tNm9Jl>iZAEr8N*Y zZZT}{R7hZ5ooPI6)NjBLhuhAMB*Q2bO=D~$lE@HSX0E{RDUA-Exi0p#-;HaA!U>Q^4jx!fsrQ0lH11WQI~U%5bT{;C>Q%w3|InDvaX;!2SJ06m{RXesK40I) zK${AEhiPwpH_}4sK5w_FE=~Ogga54{@+yo)Y>s`4eZ!IOeU3BW>>yT=@u{>)_5i2n z!c%p2KB|CS1xYdftWOx%U17;*cjBG4lkj*1$1#z4!aZ8qk$*w0{O;uX3Qm4jT{hF@ z^>W|6I0GLhueXY*@BNNQJ=I#CEM)RuMGQ%5Qzc(tJ1o1+dinEoc31=)pQXJyInjZx zBfyuk5Yu5lslSg_9wQx3Z0G;^giJnpLMQL-g{q0@SBPDuB82P0yn)vvpKiLdp=k`z zQmha*zp|I4CcPa&7Z_r9%dCWJJobcZu*M+#^)epSjMI;4X($GsGMV6MBzMw0xj)u-E_0~xbhfs+gRu#2r@ zYHD)5byqi*$pVUySLng!*+coTIg&S7Jz2yG2NY2u5sX51zg80y72baCKS$x8`-=`) zrnK>X_cl7g;>li&l#U5G9h~=(H~IQ~_qz6u@(px4i3v?J-fCl7xaqYTyL&k0On!PY zd73)RyDAFZ8l#q3614uv!Y4D?QxNvWN4|XD02rl10-?Hef!s15-p};L9Pt|9Q6wpS z0E9~LH6a?TT8enTBpLaSKLn1v&Q)hD&z~8P^1c3<*RX1IGLaK(MTo)|=$eS{h4RO) z>9-~lVjM}KGB_--n@-NM9x!#hPU23#gaCI0G)+b6Ln_$?UGe5)35Vxf4UEo>nCCyuoH3<(qz0s)S&u$kV1 zlzx2JWWrsoAmHDTV*?cU8D&pQG(o(-FE}Ma)Z(*FYiEZ3)w5|rIvI2KOl>J|@2sH} z=4TshvBtP9uBi>-2qeNmE|ApgqY3S%Zb!6}ZGs*q^Qhd@uX@#?np*AyNIKr^uP1`} zJ_(xiv3kK%`RHRxJXV~qS$`_eRkb6-nbk?iv14(H_+nSOWQ*u$C8s4OaGUK70!!=Z zt7$%B*e|1JNk~f392`sh)WQM~W}@imT;w43Jj@%p{-#G5pJgc*DVP8Xx@dkMe^dS$ zy+*Hp6Ug9T%6IDEI9yCiVls<}%)mlqsBX%QB^=aPogA zIaCpP2F?*cuM9FR;tMx%>_b3*pHL*Fx6EMM5YcrU4c`29|91|MK3Gy=Je%9RN6+0q z3lh`?Pr*A_0^l3{il16ttqeB4tG^PKNoY6xEnE$j^Abs(zYvaAqLByEkZPoYavG{x zIlnTK$iw2Xf|%XXUX3GDzXIfGJuE} z1&p7OP?3TASk;cD^MoVdt4-nbsV?M5Yu%c9t{K7s8BaI7B8HGonm9F4Ybrtb{=fl& zmG}2hih?*vddxIpY7g!_r?2w$Yi;aW)V1%A{RybfeI7rhXJi4ti(>5y@9c{po*`HB zpZER)-O*YqG*K%n|3)K_Vwin*Qcf#>q#4TSoZhqp6)T(UQc{B@-kai^-!hLiJU=X%3~ z#NV8*%50`fBF2_BW?#}|mPs>wLGBxT@n;tyz%Mk;`d*S2|Dn?&^}+%P?4q7Ro4#lQ z*Ig}42+=p9VGA~BDG9#c)$gT<;D*!;CjCXY@zlE(P?G199EjYTNTUJytjQ%er-7p@ zT72CIYV!9?w`MkHYBKDNUMqHKHq2jyA$V=*jzJwlYl_t?)%B%MvRJ61BooIZN*8;k zpX?5e;t~Om{(ZLBm>JK7u2X>X%w9pNVyYy8?w2H~z2`{@Nvj=4gn zL%bvqqlFd}N98AR`W*>~jT6aHedtJG!+)mw1A>jVyb2+-b+I8NH(v-o;$;Oa zJEkn@ap#$)>Nam7PRT~6x=!JNn04e)+gP_(K2$~kxmdI?5h1Y!NI*`gfVKSb&R^?; zCwve`16<_S$8C|3%A1aKWJ+}cuqUVpgvgJz5L`e61a^Rbzx(`)w$nge9&XzXBHcqx1Un zGzOO7PgaU_uFC&e>7R4ktGices(+|MAH^X={HfhWnf5~KC-Hz={K;F@DvPWUBfv-u zW^d#R#PXhTHXAEpi4wr`SnxR&GUF$*JZ3K4_$k8JEA`s4@KYehsBS1SDs5ukRpT^D zw~XC2@}1^gwBYY(j?CN_J=9y;t+OtG#g7&}kWJb2aWO!6+l#@wJ|Wj;WiReLMt6X4 zx_W&?&E44R;Bb!UutuMUV!!av(Lh#WF;j{v#!Ayy>UN$0u1v=5;zN*zQLHmRKyrAlgXm8q%T)Zhh4+JF0g2wqkd0eYb<>&qe!x}&D zMC!7A_nyCKd1x|_a(2JWlDYWU!?%h`;}z38Juv_8t$xJV7s0O00xBm&h7WEi9BEjsM5LgZi`^%WVHnFF%uE{Q5g%TTQHb7O-Ik#hJNZh(&n(rEPbx1XYb z{j41)G^lKk(uB8}vt z8$S2Q_EK&qJ0%XTz_)2Wt@4|rampmPyfefDDl%JC>Pk=fIs{*k7S=Ry9*2YBH*|dt zyMy;j4%-KBA;-KSE3R@|3}$dN5!=z?ZeQJ1OfyN1PE@yB}j0B1ZmtI z0t9z=3GVI|q;U=IBzTa9#@*fB;n&H%Gjr#@Gn4OI>nzsc51KxE*M6$%si*2N`?;Q5 z2A|1wX!5--H5py2Bxz0E?2!-N!Tws0b|Zc`ZdB-tP1L%j7tQTA z97ebFx4!>uNo!jF@(Xvxhza4L0~o?N%YU3}ZHNP+oy|)W!52H!T@2;H=mP!yd(aT( zC+O*+zoQ9|4qp-U(zi2&5Ck-rdu- z8x8fTyll*WOW}e@r-`b`JU8w2@W7q?6K|kAw}1ezvnEZdj`HKn8{^RRClAqXq1L(n z&MtWLp`&wtm@h;UD1PRMy_n9&Kw6AGIW<{SMd?1q?g|gUAVIP_$|4l9HcNmw>kc*< zzf$pF2?3OZ13Vy0!9Q?UcuJa%DzWzIaq0 znV`HZfImKn3jJK_M7_cb!vXwYTO|O7% z!0GSp-GopP>MvQ-&jwXgX_#)h@P}S29(-Bw;Th{|j|p?1pC}|q@+C_))Ctz{8quOI ztI1EAF-Gq87mC#k+4b}EeS42&&B!UqLvV7abd+prJKc?fm!wFz*8?*3xUrT(p^wL zR=kdBIM}rMifu7aI2I$Y1P?f_wwC&|zK;><%?sG=Ut!1KIzMYaVNp3j zPpZVLcd4#g49!Sj5?xPs{VAr0UXV!dfL45p(8=^JAzh}A;Q$gPje}3!{4jCUSYnnN ze>kon93Rmx?`@@PdZyqZUmlG~+?2Cxu=gfkr9>R2Z^2?06U_uf^k(?HjnBDGtLk+H zE9Rln!u9np{Lp)^`;F^(i}iaQg+nLr1)tiic>5c;yuoixx3Ev_ZX-w%jLxpGfHUTy zKfli(*NU(}>5-e~^DQASrtH|^)UGE#X=h?zf%)k6G&bDLg+t})XFpRhXpz?M1926a zeTF(?g7|C^21BDTtZtX#(u5s_qP7jRHgYVG0N0z0qRO~@TH7x@NxcaB*6SqB-fNr% z-wlQWBrUOn!6YT0t_HYju-l19Ci{gsP^rH6JH*Z&ei*4mSjUi_$Y$F}o1<6O@|b`q zy=(6Y6=SMAbHx#^(Z`2$UCcWWq^WnGE4b z?Q=k9U!#y*j715cRQ_;vDF`v-SLOKWpWp?Fy`JelqA#XFCDPhWvnq*ij6hAW8W&d= z8~H{)RT4KldM+RS&xtSs06lRSUkJe+b~cs?`0M#>8cAb_Jp=SVr1IElY-S}^+h1XK z8iwJ0E^!A>Rj->+xG5t>a-sE-bj!qzQ84K=PmZ%Ym+vmZY*}0EM5O-t=s~Z2g!Hb? zHZ65y=gPP2Hnv4zQTkFv!?K_ENgu9UcfI|jeGI^8pj5;B!?z+MZq12G`7TE}zHf7eqFhYK4f|@F}t{@S8mV| zN{M8+YY8JO7N#`~`E%8V8!-_xGF@+L1B^830_O18-h--N*|pi?Scf`~L$kGNM8$oi zf=wqY6Ak=V?x}T@l?$2opPi8xH)iuA2}2GSkx??;`BDEQtbYDOff@AS)Sba5@3v1k z3DEX(aZTAwvX8MpiKPR0dnY$sn{dk*QWx?yRX4NAChBx93@Uq3xiW$JZ6gGyBtp{^b%g z1>mvEcep04#fN3O#B7lw##?w z`rgOk@!f4x9+K=e)=!G?ub)Wlb4FqhSNP(AX1**4o(0)d3%Z6j8ygwq9fBFaVC~5eO(kEpueVlsptU0Mh*(4pCa56Fu z6R!`xKQ43#e2W-zRT|Cb8(!zK!>idAf>!dsbbkrE#v1KS)kY3$nT&&0!u@1)cbwV@ ztZpVwg9b*zVq`;u>5N)n?!FGz1$tPF$kdq$2S(K>%2na zs*>d^nR(S7a?URsZbqzr6ypi-Sl$!)$o%O>)tF$V`B!t16JQ}B%na2HEX&G*zua(~@u$a0!sQ|UVNL~%f zzZl<7^M?h*s=!<}uk%IW-~(hd*Z(AGg4lidDGY>@4>puPumcF2AZZ`y|F*-f=uEiJ z&#@21z#@IzF|I6SoKpkPwsnC%z(%z|4b*8YJai{(v=Bd%5dyy&*yPE@w!$aez0l+u z<9wK4Dd!N}xO_v7W&$|Zd1Cmmrma#m^0{!M{$3wO#FLT~oja*x$KZ{I+QWM&1 z(+q$6_dn2_U7s`E0WX)EZK;m7ULI`!N-=$m(cjs=E+EUIY1Ou}T&|bgFv(v63wiLO%M)p;tkOsWq~ z_#^rFr$?9fYp=uXE5 z_+@7;2qjl=1$@Jt8y&u#Fw{*7L}eSO$C7R-p+++)>)xQ`fhu^&!yH?C4cQYN?s*4SVL>#Un>X+=&_uTAp6OiU{FoIDN=ADE{>G<6 zyZd6>q0IR<@7wYP_!ivr`scbs+ZP^KD(6fZ8xq)6;DjtW^5uSIn*Z99C{K9S zxtIX0ufS^Tuf-QnJZDU;2DuI?md_hM5H~M4s%1$8vyiPFJLV>zoBK<#QCu5< z19QlM+GbeS7HfR34G+=B<9i|){EM#nNAUG)(P03ojb{)UmQnfeH`e5!@Qozb+}pjWT%;YvhlH~mlE@RPe6cnW`6E=W0BoF*CqiYlo4u% z_E7xIJH(NH$;ccFznN*fM=dXwOyW^tnS=8#q za%BO94=&d3|Jtgb#|Ihy>Ch7{$ zH=Z`;+12fnqS6L_CGZFE3`9@qW^|<|j~p9;@xLTajE?~D4f8Lfvl$w z+A~*RY0yT3)4}Q@-RU;O`^>1&DA5#Mm0Ej);k7a0C@&7QmI94n{u*3CU(59jl(V}o zOAG5UJV)EZ$JrUQnSO9mlARp<&5TaBn~nB`Pf zDg9t~`1)mYXw>k4SWuS;AV^y-+XE4{VQhKASxb1jg*Y4#=*fkLd^LIkCCM3R%myRZ z`zgAyya0m?31A6!WYdB&;~x}4-XQ!zdiRG{vX{uv&mu>A?&FTZGGqxqli%tr)5H)r z!}4RxNwuwiA^Srh?Yb4a<}_`p2a==xggmJJl>p!7SdT=NXo;!J4+rXEO(D~Xv?Hr3 zHR+JfA~^HFfAUaFN%Z+IrSSU?I&XUZ3G73^>~%4(nyd z3hn=B$B}sl#gug%&6sUmxi(R7Gh))u2y< zoIspy>fO>0W>xaz#5+DI?KvtzlgV%VsGDuDj<&%Y8N;$e1O79^i1t{h`ZpZ9iPk96 zsXutmW?6{#>ESG4XC|TL#tYBlDb2*49lGMlqx1Q+8wJ4(dG zpt8z;{iy&{I=SSuy|rl@wEol|K3j6Bi;)6r8f)>xSkq9#LTjV-Q0qMMSd{fgN?cd` zBYO)43FpT2l{ckZRU2G0T5!nC<7ppClf6^xaAYjc_gH0-=~99MDd*>ma2(oLfXJbE zZ76&4RE{JeuH{gz?7L!vpqKSn_^c{nLV3xUi>+XmMwWbeAao4ol^%+$QEt`TXDSM0 z1Ce|_v_Yt*XS6Ytu~#3rq3Q?G>eG>U`m`*h^Z74Xir_WX7i_OfW1%l?LA-wKz6yYbx7!Hj zv06?qKb?p>Ef#5N6RIA?)nBs@vOlK!KK>~c`(;7ol(5sBzG}85k^={oD%9e-H>mCO z<=lZ6qB7G#O1*`sB00;^W7gf|O}FA#xpR4OM-FNk?2}nwNu8$LWMFRHh2JasPiCt_ zODE6UpAB~5hxK>2Ft32>6E$bp_pHmwlH#f)VESi8CNyq_T?A%P2U}!)%N)^UnkY(8zZ>7OvbYAuF{Hwg^|EPLAeyBFvU!@pZK$e=ziAX z_ahB{4ce(6FV2D@_5Kg4=#K>#qXf)?sxa54;w94!brihYXWD#dJksE{9eIaOKjibN zh7b9#G0`2F@tzOC^O?Og5BJ%0dg~6s`;J2W(wQ-a-^eYwAMecH<#l`r5TS!%Bb;N$ zC$B#D37HgmnijO+xn_rX1TrU8>`Qp^j-gv+f#uBc7|XGyj|LuqG{b7vNGsZpN-)y)0KLa&t^L9 z=!1TilfP=7vTT(T1}uu#u3l zERZK|M^5quTilnwLd+lcouTdQCe9_VpRJC=)~0sDk7bY-f4hygtWG%=yVB*;Mqzzr zQD(G&**+c*ZTA>XK}O;^haB|@eSh2y85B5wzrphm7Kv?vX4Tm-u-;r#Y9?MF!buYf z8=tzz1JiuRHtZ1as^mVg@NiHdztNK7^JJ^i3MM^qd(eG-{@iU)5QXKPjU7T*` z1bLg#!_HsfP~$ehIIqCjS(y4|cfWyhBeFZ3y1RUvN!cD+Czi+Z4KL`g=QZ?O_!(|= zkt-x8kJRJ6uS?S%0*e~{fNMylf#D%6O6H#MKNb1^giRar3THSbPh2MXurCxKI=bRA0=>EbUoe}6L2VU_yuQe+@&FNX^ zJ`LlPssU>~voCwL*0-ENMM(XCoQbEBD3B|)GD2I6OAmP9ahI}yopaQ#R=+59igA`R zH9yK`-BMmkHgi91Oef%?@sin3FKpS>@Uj|;8)kKCd(zzWT{9s-#*pEce*G_w6UtX% zydybXC})5MeQ;y(ml;Cs=*g@hiNonm{~`V!qasH2LhA=SN9;^)hfagFRd?FF>yS{B zZ2KZ8+JbV15 z?Rw*Ap-Q;rM*l>2+J&3b7fp0KbyDMd7x%^@(xdg*;9=kei6%fl0R~n2OsC&l!~4SF zZZp_mhd#$=yEvU~D4KJcODWB@DX(l!V&HmQ#ciRm-HG65V55W&Q*tw^*6*7LW5w$D z-|;Vq$>&}V`MfW(z08+Ign`@lAt61v2^F@8pU!Thr&B0Sn!V7N$C>x55uB~TOQ+S9 zBJKI{VxSH0$p-w%ZgfA_^N`d_fNeq3C5wMt|2!apj z>+%jH%e$@=5;g11UgLYX#Gafl+mY`pI{4|bAJ8}?Dy;F;3%2^UckaiDqrmy-fXUdO zWn=QnCZaEPqLgYCFACVcCue_LVb#<48!NWqSOV>xxp@$5@*C3-Pegq8d z8Hhqc2r#<7r$K}wBSwslu*Zw6$*Czwes6NV z(ShAwswaz4bk|7WY(`!2LEEb6`J63-EV)%d%@ds=dcU{{*7f(y37q^_T%%c?U#0Z0 z*rzxlY~Wz0WNpeHJ~&07uTn9}xD<22Y&N|L zDatQUXyQb{I&aLItP=}y9a2DbrBr(ea|Y9zNd$*S1jtk_4=De2i~sBW52C779^CPm`T7&6v@rnz?flUQffI$R#Lf3q>!V`sC-Q0F`N;mb^q&jawtv zz0qzAU%QQ1nch;Y8{67#n@8p$n=ElL>(bA7m0rZGql_bCsK3rN@BX<;1}en6RCLb5O8k`jO7UY|GCXRkC;=$hYL54xQoz8M>w*b_N(k;65Uj)u#i|hovuXI z(4Z_iR!9D?m`o&K`~9)s=*JYlSq&I5Rz9K;@CjJ^-~%aNG71%=c>2!!ALNsnPuy<` zsQnv-shH%_XculBbz7QCcq)cOYz{vQ|opX?OD+q+RRi9mqQ5E{oKBtm5|371%IygTUzna49CssjH!gR~Q zb~32>isEPu3(Bp6pYpkMLP^OLjNwD;c)oUYhw4|}!W|0oFoOYcpY|6grX>BzEP`^M z2D)f9*Ydh^9_?u5X?IFPxsa(!L*PGSjQ@2*4Bx4Rs7%fpJlF(-QfDXi@{539%Lnqa z_UVQ=vSAO(VY37~)x?Y!bf&9g3i{~pba=Ym+qx3F*Yeu6aBhA+Ob1m<#DK?K#@i!1 zHti9xT!-s*>=PM)it<@vhu2KSH3qg#5hr+X6J}K7XT?MkK-EbxRYO)9W#jV)e?K8$ z*Tb<$pR>;0Ff$#v+0~@J=_>8zOpZdmY-04AoJezZahX4 zyDBoUDvC$Y*L6z^IDVBQs$P?~&)uQrrj+vGx-$7+>QuJ==OWuh~($ z2}@MVlrNW>1fS9h*L*LvAHne|v0eE3anjF#`3mXFyy#Gw#SUzm^3=~pDU5W19Om|C#c3^f zcg$Z7aG|<0Y}i=XQ5xpwoH}7bcjCoPK5|HdoHW>Q(0sMJ6eW%pu-0#~KI?&S+e?ymM1MU_Vv()5{fA7|4>_B(a67f%x`SSDlN;#;disQccxpWUBNp7LQ3mI3`b zIM$hX@RW-#fEYBU=?#kF%vjgObx?GiiG_$I zsIp?^uwM{95V;w_GANy&FK5Qti*W=M1W6Q5`OME3OqF_PlM)wtKcTFBqeqLK^d^b; z=@(L(7RQ7^yABAj{ zsK6^U9*OcCkY~?Rw<%PrC92oSBMPR;}W`PgKPX@1Z=RtLh{Z{?jt{B#b z#q`foqD;ikPMdDY3pE0_W;f;gU+(4>&EKmui)xwdls-6g?TcI~&UNCV{?7-4J}z+& z;=7SpOny+{{a0C2BkDBvsVA#qj;9|Xyi}q^Qq7MZNI?#BtI@ZEW4*Kt=}XesoMxd& zAI~%Q<$)!f3vNO}kEMW4i^iL==oZpM?d=&M}REiHDoK4B)Ei z>|TQ7Al3&kkIV`oZL;%zD1NRo_C9sA@{D$%<$~vhI@NaAgeTh-B)Hu>a_^jAwP%+r z)L60Bo~xA9V`GK1t^vW)ll@9~JE7y=G-NUGsPHQJE$M&Vv_H2}KpIokyx5D*0^zk4 zAcvWfm8Mry@fprM$uzrD@|e4KwGHqanTKnsz3S=b*R@8QbTp z!I9pWm5MD=>Q&6c;Dd|T4-b@ul7&|{gZ??(syqE)Lt*W0(5-|_I%Wf3x#8NSd=I({ z=!&R`h-BVHa797q;W)DTG(CEg$^aW+o?8TSIGlnjqM3v_m{ zK2j}N&fd`)SLQO$ozFr6YP*qO6K~t_$ILf6xf3w`hW9e}R$6?{-<9gQwa25W97y|N zB-?Q?ddQ4+Usb3`Wg1>(9i*=@rQh%lCn~!`C3Pw*29fZ7pz`jlajPqe;eKlWP+X|{ z`dqmJ3?gsV^V*gSt*tg{`8?aLFzXU`)_&E?YmQ4AKgD3n%T`Dws~X4BhKt=}l)6}- zyfoY8=iQb{%oGE{Vba-6@1cAP@I$`p(B-($WJtj$IT}X(fH~Ne7A3T@S}3 zvzw&M4?CeOp21BGkB#ldQfnI%3-O0=Ql@=_#I`HD`7|4$ zH+8FF<)Y>T8N#($9Y4@++l0QU@kDh!wxv$)_cpEYL z$o}5nLfV?sxnyjhQ;#48__72{Ly2V<-I%tt-Ruat=)y`bW*U*?3g*#UPN%ilOpd6R z%bZf?&M{hQz?qe67-@VAgb*BxYVRg;NIYW=*VLI~L&ywCeYg7le}fRzo(0Upd_r%k zsp(Z;KLm+bUw3fFk6y$r)OYU6#2F2+Zrft=(yQWXM5?I6C-n@4>?bMKdEU`(7fq6a z^_CDJ775)*Kt@1Sf*7iYT(z;tH`?niVXLNb#EXhd2Kcms(p*FyvSi@tHpD}38@OcZ zX5}%kpr2FIBoou<5LD}6S))-q9l^aDuxk_Np|n+D_`R(Yd*pT}+F+GKKKwOVB^*CbXG?Jd0Tz(XAhf-1B6L|YpSYtr_7sjy4$PuXe5Rx6^?qTd8o zRFYkI6PP|Kyg#f@KJ$!qIa|#TFZHGBwYZ6^g(xt3toCpS3qWrL_m;7H%`_NNObgjATguic1vE#nN8}t68<_pGoX` zkba`uq8cwvtN*6S8j+C$y6x%0dA~VZ27ykF8__D%`?dP$1H$G^a-`5 zub><=U~wl#Qk^Cjg!JL=!pZzTq3(1YGI?Lkgl4f3luqxej=P z7tkKM2;7Ve;;9AlZ$o7Dw#C~;+)RhLZyNU1HEwFWSuIB{U6{U3sOG;iu zDvoy3E0ydKz38vBLmnMO;{(m&e&qYD`Kd{_Bc@T#T%=nhz)8KeNHxm&YwcX5N!D1-+hsZKe`~tyV-?;*3bA zO@VSFjAx3Tp;rl<0a}_Z3e&8Yqh4jcG6h%kzTu6Gxgo;o7ps<{^r`cTk+x|Z1?)Ki z<))hrVNnVssC~2!L$oXEJ*Z7e5ha0Z&)!LF&yAWKEj~s&HoUd^@KhpiOHpW+u5;Au z!W)spN=#vy@_T-BRVUf+TpbL2jD&9J`9rBi=4>sWF?phaNs&eWXZOBEf=@yj7hz6m z_bFmJ^TvHv7}hq<{mR=>fhBEf3pODu$qs>I27%eH)tY1kR}W>i zQ*1*;0*mmpZZ%D7zPCmwwhUTzNsn)L%7ojm98jDOee$$`bOatz^A5pP{f}7fkK@{F z3J5iKa0F>Z+nO|J+DYG4Lu%J%P5TS#QbF~0)UBUObA!`cTf|LgiKjHGJrX}0?k!_a znre?kpz{CFs@K*V;f2f-vS?{c*5wXT-h^~-oGxM$PU#r`PQI@(HbLL~!d2I+k<5{E zIFRVj(T)B&5AXXOJ)rg{K(X-v2#HqmP27Q8G<%Zc{>)@5aKQX{==+aFaK4^O&#I zt-)RpSkAA-!(cmso4|fF+b{y&!U57(You?@j(#Tbctc^0z0g|!L9@k{S zX(^R1-%?IH$e!L(j=oZ6c(VTuRf6G@F1g#*nZQ`nool|gX`GH^EU%x8i7jc;)89dX zKrMQb6_V=&J5C?9a@Em$+YrP$k>t1ZAtajW(3hp$;YVz}ahU6GW56rNOwzphy2c%rKd{apZsaoG{1F zpVdY%pHNi70uO7BePU*xtH%-kd;XNO$n1>Bjcp+>QdJ9{gphRn3;US})>`_D=;4aU z>Z7)Sg7acAuXCwYM5>Rb>DqA_Z;w?zUToJuJ`N(8#k-y3@6&c?_|kW;@kdW;XN6VY zShn5F52|&PlYZj!lc5#@xBu4sr-AMM*e(OLk7bfLE?Epav062WOvixN7zU|JIVejQx za&+CpBx8cKRlLdyb5grU<`cd;UytQme0n{t?U-e46^2h5+_OI#1l3=C&#!)mA_@3y zlGlWwT7cWY>J-dM_1ShgZerHZH0RM`2q(R?x+4WO>Y>)&e)%FRI5dLhbG$O@ z6HwF(nO>i)Vw!Y$Z%ikps^n-jZ%z+gX6NMs!vwZ8+N4P?Yj%{F!nb?E)~d5Uwr_+N zm{RSaoY&0aw$bf)it&!@$64Hu=SAeMxoekBS7cd-ywBA9iojN45B1wD--+_z7Osar z8`;gGL)tN)TgDdG;M2`0G{3D4Av*gRw1XDjCZr0Dj!E8o~rZLwQce*JaH2TJDrsS<1cfzAtH{hIGiZzSR3+% z4RM}z1j!7EK{-kGPuZ{g{pbk@YlGAuKV&om5gjkZDb>wz8l-sGq@9%-&CGe{oI=LV zho>U491bw9Og<8es4@9i5&*!vx;pHKb!}}+Vs2E`aSN>j!c5y)ed0|D%MdT^qtB1k zDeCamR*hmcwAYh?uKT`HZh?*Ris}uvAjHhtFo}^^y_AZ=P zhN(cQM81t|p7De2^hn_8nc3x3Hrk*%jh3-oMr+#oJ{YVp_;neXy|VAqZOT#4rr;TK zq&QJZQZHB;Iff?jyqMZ^@n#U#XJXi9B>&9LjL0R+-aK?A(~2Q{i_|Lp?VPrQ*`113 zk)QqUvQM?+Y~qO(X8B|=EU=5%)l^Oc&%=ML+O?U?&Yc0>n))_RJ+4_d)w8J@SgI>? zap1+RV*5S+v-!GiA>`soP49lA5K&C`oswp#5H9rpe;UEx%sS$9KL#tF{2nBSRcRgENVq&y@GKk;AsD5k4XsRX% zlHzfO?9f>24s$_V%PZ<^XAS$s=_gVPk#qg$M z-_4;$BFa6@Ks{fNr-_Ru{JMG3wSUUf=hEy}0vdMXXlzBD8G#UL6SI|IYq_bvCr)Yy z+&Y`ga432=81b4&AV+t9zU|NTz$vZh=z!z`gd;LF$p~8QC`)PB>`-v{I2#~uH>Zv6 z_>KoQF}UkL@(^@-(0$LZM~A@v-n3uCD*V{muNJat5J7aIp_2$oUBezVj(h z**{Ev_oGu>gPvFV;|4s7v^k{N@FV@{ZswJG+Z^N} zX^9qlwyqLUgU(?oVqBk-=6)t1pHNct{WnxfZ)2m8jJp9|qlKpLq`_)j4F=QDM(B;~ z!>|p8i_oLJ-vq@tJ`#^JjkTJ6 zb>PkU1vJC=!+19$KhbJLZ8pF8c%LuEL5QJFh2Hf1nZabKaGI;c57tjTHVo{mfWG84-LTA_~n|nRb?z6TxeO63EI-~<+)Zl~I zFU8s(R~PlNco%mSFUo;_m!Zj!Yp^H|Y-wTel4e}~eG0jr!A-9cfMG=rq#>lDGq~Wi zIcvw8((9TfYAD@@XbE(rc5lTj9pt0r>B^>BBE4wbiC<&+aTjwK+o@TfnAW)IZYY+7p6l^fOiL zBVQWJ;$Cu^;F>d*$L-PRnbp zF5teZLR4J-qZIN3qs$%}kqKZrgJx=z0lC6{)hd7Bm93$32B^kuJ8CuS&_Et|bea!vC7D43|Pc(d!=UeA%((OHVXaOK366SmUivFP3Z;vpV z7>y5A>E#n~0+NX#qF0O#gzLOJ_@TrnTZxoQ*Fzeco6}j%pwHYN8fUVn;%Js9vyEFn z%i~h)et-CJ4k>dIM^fqLGr|e}QE9imsmF{{^J&itQc<4Ek>71=uvWWn{vz}i7>^Mr zK@VxLj*h#q#Bw3*QaU|ggoqzdU>nXh8LM6faAL~Z18M?VKvHtKHAsXEYT(%=<|!}% zfZ=;`s6DnhID^VzhH5I7_tr|1nPk)&NVUDdnp1J@0diH!HWrH|$EaXLpvglEM^}E| znUA1r{aYeWfS4W7QQF`yBSGeNOSK)BAIe8UW!tFdRk&Cw?|?H#@(o+ud^JG4+S|Yv z6ykPg)(4uT4QX-CvbUU_ikQ>OPFC@Z(_{q|iVt`4no^Iay`o3oaAGi3X7y3!FEjvj zS&F2a1&Zc^AC3s`FJbxy-$q}2Mg%Z=h|fTsy!K;*LQ%Beq_}z27l&H3q*}2VkEj-$ zr}PEtoWKIpa&ap60ruN9r)@tf@=I-kQi>Zp(is~Q9LaWSuq1k#AhhZxc!iAe+kiw0 zexai4E=?IJrApFN51lzgkA7E-Av_9bl}N|!<#)Dnxv&5&gBqTl-Gr{FRmnAOmwIne zPplTJvVvfHfRjn}1)Yu~MEicvefgss+i>ogMNGDZ(SXme*L#QcC-8Ef*ZJLOZm(QG zeW~kx*M9IBPwS-tcU>TbCA;1Jx$QbF_QxJbXqTz4bettG&qR(fs!uOGw7%f#DrkI< z^6VKTJizsB8|%kf`?Va1sT*TR7aCueAg!7OzG_b*W$|gjjmQ<&OG>=o=Ney_014dE z*fUgv1B@{;9~GKcjSI^~TqJ7GU(wH>S3bTu)*g1G-e+)r@0uT^UTrrccA0K!5=V0j zW{hzZpj!tfcHM3Fz?8ANr-;`_PsypSe6Bse~=w}`bLkHOGn|U`EBek5P zIaW^I$6)>dLDH7WT%Amkd3DkT_w~1LiXN;I`Eft!UdR^~ia0)J&u_WAW;fx$1R2Ik zK?hHC5Ds}Q1e6bgX3|(W=J1z}LXX?!rM{#~E&0Vs1C|vej^DofUBn`AftF+hbfMf& zSoM2naS&{&&ZzY!{nO`6)ogr`$ttg6d^)iSA)Twqmr7zR9+j+nwA3?UhA1qoveOA_5Pg&0jUL!Q1#a< zqxY!6#IlbsOuqdX&cHZ#Mcrqs0cB#RELC|=4rN?;-X$P1Iqe;$Z;GmD#1&{Yd-8Ni zwevC6h}xjyu@HXGM|+L+->T35Iw-P=0EUH-LgebA0s`{=`X{uxLiE|)M2}af%xhTN z-9l2t%R1?-hpRfNNt@%n^WfUh$8e`0%}kZIX2JIRyO7(q4VRFMtF2`0dt?4W#o(!k z$4}Ard-{;(b$_Bw`DfY?ZLY=724h8v&bQ6jzVt~`7Ka=0fJjFBYR`-3=UwV}kNLbD zgeUwo=p(szZV`jKR9g+!R@)7y`ny#1=cJhNb0-n3D^0pqBNuH=;--@o#yylf1)Egj zIN@g(7t;=oo=zt>$I#yLm%E@MzUZ)wcicx0ajkD7EewgI6C)}w2gPd5=uPI%A%|YT zXSdCwr48K7=!>M`MCFSfWU8&oMVityYI8(o$I^T|t3dVsR12U*J+yu(m&t@D_`CXt z0Q8t;s{104ggxi3PTENq+u~@7nw!afG9Jtvu92NzKO)~~VU->IeXp7q6oRd^xt~e* zH6j?0##VzGS1JC>(AZd2Lfxoly_5JaxyZZ=OhB1>p|R zXcX2(%2bj2@%EwS(%S7Y#Ngn5J~(4XDVne{a%^`2!6LaJ$SSLaGOv2Dk$Pl$c5))C zi;>}Ch;$@_w?qi%n7RhaQ`X$!68x4wDn=-&9p}-V-6&C4V!;`;@nb!?HHcVRbW*~U z;`}IMG&|=hw>K0C%E&?D`)W?T9cu=HOTnHit59z%5Iufwxxr9{<#vCsaB9)Wf6=Sr zUz+viyed&b>eeMv(=Af5ta_5Duc~Vdfl*a4mwO>`t}%x>@x$)>ro1*Yi$2Dcgx#DJ zUZ;c?VR?0H)xqS*{rL0?n_}AhJlLo{HTZDg%eHw@@u+hii5)M^C^BEC^s5WB&?PKt z!3AkbSKH_xc#0@=Cy+$C;Jwb_Cz+$lwuNe=gSF;0mxE;wRi?(f$+JCe40bIlGh@5C z?l%aT@}AF={%fZE50Qv~Bl;_|{%e+|gt*S}`ZpgKQsiuA(t9*an|@p|ei2Ey`~=@=3+Z{5{Qejb}F4vqz#OV#q4?2X9?oxuN(bug(=+ zw#On(bH2TCWs}s(nSD|awEos*qKom{Ky6y6UC?n$ZdTa<923(V6X_Wn5fSx)E?TIr z$S9^7AI{`-WxGz@$>h4Rm$%tA+Fm)e_q=kYdq? zhWbh@WkrF>8fm2R8D8STQCxmY4-O@8wEicq+3F| z1d;A;MiGP&kZw?9=x!KFx{;w7QaXl)A&2vT@4NT+oxRU_zrQ-eJkNcvd#!6->ssp$ zNz+e)%JN>iS;p&rukzt9A&U^Q&{7BPnwKaBpqG}3e zQlw8QTUqB47GfyMIn_0Jki;Wf#wR~X$cFP^L7glGJ?n4N`hVKqS6lS=lOHlLufE#$ zhMNjU(ww&qEnndZcYjdRFKu$M!OoSefqlbF6}9?VU*s#i4^p>}JkBAl6^TnK`T-n) zlV7y@hw+?tUv- zAzwXn$Vjcr5C}yXC2RS-s!E zy@dOHq*w1;vcusCV`h7P=FBFlbK=Z}hJn$-sSiA?cH|d%@?t6(vZchW1Z+!E3TD>r z{|>?b`wjuy(lPm=MT2#+stvuk6YCBK*<<{oHn?jwF08+5H1CnU%4sKbjgpw=^9OzT zQc{kRv*)LX#Yu0s3@i%6{o~y-Tcr3%F?9%S`LnTu6KYT^1R#(rJkAas)o zeede_WPKU@!Mxi~6|4U2hs!4PZuWattGRa1>CfS6oeiPRZ5O3+iDCtzexUSdAwv_ATuH{JWRQ@cW8fMvvizD5cl1kRlX!Z9?1r%YxoW&T2TaRR zFY{n^WzULnz3;GF?9H^rLxEXtLWqKIMiITsSCfq`Ed|nQYlI2eR@2)t+p;y62!-_i zUCME(BJWER?;SRT(N4WYhi7}1GTV#esjkik=kj$cm0d^eI>YZU2)AbG-W0!1`N3da zATU;X9w19=w&kxNbrMtHtladEVFLez62yLtG=*=m2*j15XC*$$g+x5EtbkYu3~qfzP!}ame@x}G5Bh-p zPe|<+S_mUoZ{1T{cvPFaQiL4iZreC4w%99p^0}l_?(joPj-D@IUP2oV-?s= znUvus*n(G{>#>xVG7T1#>+8-Li|eE_eGMzq)`NpI4Jyzoj?r{NNbIxHH?y@aE!>Ry z=k8&LQ%kqTP4#5o^vWqqE^p{RaW6`#Jy6?dTK!q=3>2XoI0dQWlGP;^)7Az*H0R`A z=yl%{o#JKmkv9UkiN@8S#DC1#-;smZt7xEgd4smtP*|#~b$`*_RMA&nd*{DiAT%V- zewgQQ&PIVBs1KmCEgW|}j%!?RcCfcGV@$I6rKfAw65=4(02G}yYD8yx@TL^Z=R4Jg zH*Pz&U^Ly6#1ho4jmBa*E?}&*Hz9DJJ~~5B1Q3ktrR!?IOM3BEqA4Xlu#oGHd1NxA z+<|?+z!|#IZJ)_%xcA&+$4P-OepZ)oC$oPM{gu{~*L2-QUqi64i`O9e~bb8E*nqIBW$HeQ5;?_OMr#XJ_}P;;N*?NSw?OT5R7uD-oZ!LG=6 z>beY9FGzsPSz6;>*##lj=DcP5t-UV0cNG)uRn4}sCBN@&`hS;dmCu~$wSV9fLO)M) zA7O=-CK?}j3to8DbQk=iaL4kg2yYwE--LsnFR zXibKGy1Ieit*Y>ot>C2QR8=2Wph}`+S_1t)My;`j4$q#)O;)j39AI@vS8fh|`E+U! zaJ9>_o?}4BcjZcg$gNkYVB~vMPr7Zu6SHzBQzx#9b$a`y?uNGwAoZgFl?P$KjF)2% zDEu2{0$D5f<-5IIa$AonX=+Dw{f={_?mz_wER6jwe=Bg@jvs(QDj!s@4ptvpdIx(C zHwn$t=8iwQ@;nw*-tdg^HxQpqRs2QD@!*MH+7fK}Q}Ud5h4sPGB+TsmQ+3wU&r2FZ zVl}%ScJQB7539`5sC|0sojblz|L>UPU;hjEGDgQ=M?yXQcePrJk-&_h*comVFfW*mViQ_w}DNavMp$XZ& zqqbb8wqu%k_U0A5P1kAt(#~i2D#r~65)wC?|BfLDW8X+^z~gdO9jN&0yHvyIxv^U% zj-NLCT;<H=xRE9n1(JApmSBnG{Yb7IyX zaWWLNT)Mlb;(W*hFs~Qqr-!=hB`YY`Auq;($Yb)~wulV$S{Ga3jBV~N5!TGmIN}_J zw8hTR>>CbmfG|ZO^@#MHe@DZ-n@*~0#{dnWjpcSpAMG#4zy^1pomAvQuyTdOJt;;f zlOCYB&a21LZ_H}@s()2LezgkcxTDeNh}W|H?APLKlbS%5RL+huVWjU{lQIYBrR@5^ zVrxT8c^E+J+X((B!1(7NhA;aPq@ETxi|U3WTKv>fdkp<6((u?nO+AY^t!}iSv^?=d zxt4k)Pdzo~Jfd)nlcM#4M2|->#xw`@t4$_L_zuMdwMDLM#!^C@(+((e(}Y*E5-kct zA;YcD!OfwrYBPeFuF0=f5isSV$~BY3cjFBl!uw`xUwY&2*B1xMnBBJ}Ht6Pr8G3|v z(_3ef316HiUw59gHsRVICy@**mhk4{dqyuoH^7B$`bq6ph9yo#h1raN zq@DWf{tv`zzTSpXW#a;J<_Qisy+UY%$}t?JBMuA4?Km)yV37AP)CZ}p(2ul3J!OrF zoUI#C3)qg@0hI9!LjT%G@!3-$9Wh~1+^7Nv=*FGi$-sy5Rf%u@$)iy@dE}_p-n6rt(dTQ^1ywH0xP}H%|l+|FiF?pbRJe?Mt<85R~naL{Nme7 z^MJ#Up4`3|8Kml0mAUB^CCnM9(#@cF^`H0%P5hIw493a`eBo~P-R)}Z3j=fqbKYpg$kw$TP@`YA2m!0w>k5CQf+3uI`bK|g>HEnl|(!p-dN4KwiUnu;% zI;7)#&mQ3KXsT1-7}S#Q1lA48y8_uGKarZy5K>S_$ z-Hiy?1VTo!@wdsZs)M z!~1l4jwt=KN=tlM;W~5S^$xy+dQc~B)of{%U0?dG7Dj&#kN+Tw-Nrn5xb~tWquggn z@6Ef_NT@8Qu1gddLxx;nA)M|7f>bt^Xx6C2fvP^-$qLKGTdxjT-sL>;T#}{KV9Co0 zmfYOVDc{<3G(bR|5g&38vtjl7Tj6(CK#fw1R4|Y5XE!*qG8VnCuj6tZDj9f6A&c=y zq*J>`naEoatR3E!T7{YJw5ngO_D{G8@ak0J=-Zbgu&IXVgaNyk+D?%^51B0Z^W%l380+-K>DR0ii>Cep&0i9Fk9 zZHKY84OBioRop#Z|6F*u1u~)L@@yPgNjnK5vnMUXtHPcNsdy2`QTUr3CG6+;(Pcgy z(=1hKX+M;rPbMt7>HK8`&dcVEb9zGjxzM|KtCY z8vAcN6D8X$waULFzP>p*gjeQIs1WO3543l!MT9`i^axEaEiwumoK@BID~Q-GF@=l? z>IBXu*POUrUjH+gdZSi|z!Pli}LS+NyuD%L0t9gmQiDk_ivpnzO~HO`zB zxgfRe8x9J}kjQw?WK#iDw3ZJsL4$+2r#NNbjpp;mxtyp|9+JDRbmzXlU$~QXI#10-IBT&cGe+^_6PcacEL)13~vE!dE9m%7M@S3Ef|t66gV)!>}Ig za`A$((dV1gH+hpx<}1}b66tvy@GX~)a#Z1l-dVX3TOAnuG{{T@bkiatvM=Db#ZXy9 zSt~+nP17LaR!FaO1yAovf+~q=b~JJ5%qr7DRvAu_jeuiq$G)T5aG|&4?XPRxk`tV1 zc2_dYiJ)b`BGvzax1`T+{{J@I|7}UdqOb}o4wbKpBJ;GqvgpZ2AlPj2?%e>x5PGgf zBg@s>acDi+Z~~Ic%pX$XKVjfzX@xN0%8DbD53P@I+>qA*f!TD;#GWI~Qr6j^3I5%IH6gTddhMxFAmjcl|HS;roE79XncGW_75$lrQb}-Z^;TKzNaWKL7^U?X< zcW6czk7Z?Qtt5i~tWFI2T#{wLA2+@(*JMU#gypl&vPk}USx!f3NTa~?o2DicW zFOFK#-i8HbnYg=C$MCq?<=I3pPh9r;6`VR{X+Tnzh$=Xi=>FI+en0?^ryrGVdsAGi z4}!rm-}kE)J1as>&_U+yoAa%)POEymq`oAq6pOxI(7@r;5nNV9j5>AXK!@<>kf z555DrQ15PxQxhEN8lz`W7oEJ zl%8IAVUJIbPwP?;gfcE5bhQJ7vqBSFD;9u)m#;p%j}%z--qz`WH={7Iv}>Yyw8>UwQ5UhzYTG`FRF9-!f4($6=#Z=0?!PQREI z%Gd7j``9RMHB2?AULWySqLZ*%9c@@cH9~5QwV-a(1_)O=z+?CUVrWv1Zg)j4Cd$qID$l5U9Xz}es0W@epg9jG?%qJ7U{wL@XmOM-i_)1c11;eZ zMlvA^`I2RV@LU9FKuMtX2@NAHq`n10$JZ`b-9Um|klCf~ZtW;gmr5z6&Iy6C&JNLW z>69*XSFgyWKxXLg#jHhNnfw$2@HjYb)ovBXxih(LkKFU5z+kt3%^$iRp>aIiVkRNr z9R$){r;KcEM1xT#Vfq&zjRP`R1stMAPgDK<-d`Jpg-(M-)}w6&1$1-oj}wCd(?}F= z-Z}Eov|f)@-s^8trR=`M-7!ZzQ6K~k;8GFZg8FuvwXasc#g^Mt;EtU-Y!epPH!80OrvHKOI?B~$s3#f+33R zbaZ=SnR#iX8a20YaeLv@P2j9x+`XV%&xk|~*YDw5uH^#w4_NO*x_L&Q4ReBYuXcjH z?#p_^BTY+)RMmcY+L*illRLeP0lUoUcfEm~e27&;$NbEHs|M5yZ-RNni=eaz(W zX{n^zw)4BYND30OjQ2k;i7MST&;Y2QR_UY`y`lEQN-b4hjhL0Q99$c$H?jTr3ScuS z4Yws8{N=`%1a2b}{l`1iD8bun=Kgw_IZ4FV4&dL#A(;3Mq=7S^-ikuEsn)*_|Mv%A z>0;$5khPH?_vNZ`uJqj|G9Tbs>mFLT7az77eJ}Uk_E~Rfknk;ydCwV}bPV=s1G~{S z*G?NhMZ;WB$Hcu4eEKij`w@uAx<28#UO0*~~b;3W#n@{5!RP8Bq~+PXvL zp0^2n=Xp;?Q&@yg-`1>}DfTRA`I-b6`~dL2K=`iH(}jwgVsCrElt`9y-!Acv*E8Up zvo!hD`|b(8{M5q87u{Cz+K`6srbc*`;7F?kF_Gse**Px zihuQXfYAAUmK`LV}3-Z}zEz7Su^4KaxbLAG%|6{9mS6>Mw@ zhtyA3ta^F#yrf^g8lH$W^~dOC3U4mQtCv=rvT4&CWZ2$TnFL&`~l3 zs4n48R`u&JuA(<1-1d4J_+Q2>Hxm^lHrH(6;d5ULFO&`(v^vpO9_E3N2LEB{4Sp7r`8D}`hQ-#eg zrSo&1O=n?QAqp1qV3SeM&bQ@4*j&e=R#1pR>wZ4%8q?Pr!C zESoEAN9en-!vh-2(|yUKW6#=KzRI=)-?4?U`^-T&6F*yLxTE$0IS-G6{xFO{PniJWLw|0C zvuP3anZ-bUSkF;uB$I7M2E7}pPTp{dT}h6Hnf>v=<#u$Y9XX)M`&S-h_59R8I#yjta5XRs^|`Ys|5oU|4E&oN2cbApf!;tV61 z^K*p2*>o3TQ={w)p2%DXEoG!2Fz(?GO+c60>_W&`$rY(4US;xisZFJH;@6&9yef!x zTR?onPm)cBvAPg?%V3#z=MgfEZN?>GJmN3z^ff-{logOE%kJoC`~;>B=#xRU`ydgv zz~rWMW+=a;7Q)_g6HKeIsgbAWkeHB7$WtZ(}w-9RV z{(BYx30v#nnBG(!;zw5(*SnOF(ZI;Nk1f+^hs95b5^-Jt%ku!8q_7Me*FZ!0t0sx3 zbpw3Xj^OL{-fB6o%pv#uqA~i^RJ6v&LbN{ciE4UF$m^gRlNCkVK0rirD(>>UpmK9s zwm`%PXS%uZY3F;X^$L)%K zyY-x5+c0>%=cUJA_;Ov|7&LQ>zncd7v^seyqYA+FVk-|POlUgP;`W%L*sGhV-}Ff3 z#RmCXx8#xniRR{J`WzKvoX#m(zZmGS+8n$!hIb7%sxl-zHtS`+`g66iJB55+B)DemtgO;Lp00QGgE@8}-vG9@$BP-N75cw9^A z;zmYQ%j5U+xYGCizKR|f!STOyV4`SX&8hx)_mk0ytH(&~P+Kr6@L%8eTMd@aZobRb zVx)|)kNNsoDKHd{CVNKbn+h7K~JnSZ|b^rgm` zSJhPjX?1)W|9BGzyk0pHUK3GaX}#8i+5phv!;K$*Dk`-K%~Eppz7>Z{@?$7MXUIw$@y z?v;OA)Ab>PMT!+tz(lPda+i~{3&foJfr*s?ueR{T0#Fl4QvK@Vcs-#1+lYS`eZ7kI zVzX@2`+^XbVEH}Y`t_TnamEoDtedmy6y5x$SUw_oK9|Zl{qc4=Exhm@sik$oZa(rE zkR5aVTUZ6!0=Dyy+?^qSTu)S&E>jd0*_cQ^B?FY>=)n~N{!-0X9qn;$&^H#ziY&l% ztiyr9*~ieeO^*A5f3h*)x|~yib*poOtz}Bc54T9e@B*QNd{hITFB_3UeE)%T@s^e$L_?>pZ64*t3r@6SFi;exJbUwKIz z7Au&$Uopq^LIE}8q$Y4ADvTypO=o&wDYI(pQIl1=V}C{B8&3sV!awo+4m4fy93wN1 z{KbZw+s%;%Lc@JnHA%XeN4vnoJ8hefn@w3IZ~`U;+M?&$7tCG6o;`A$Y3oG?e*CKe z`g9nU_Z;PU5}8)CG(eOUJUK6|^#K~y^+nDL$kC+eO8Q1l1znJ$HJy>LfW^zrvKgq? z#~FV2eD_$<2^;?rq z7uMPs5&3)b{!U)I7)zXdsiL+vqapSP6Z!fv7Ys1&C_VSEa{7zuK)nv?X}{`2npziA zxsx;IYosSjiq^_FwPV70I{3aX7FNm*5ML8QFRs!=@l>W3Q? zzr|4EnePJBdmGZ_kDJ0em1iWpicG0S1GAsYJeBBFFxc8YbXW{;a-fO%SMP%_tGb#| z3}sRYsd9!-+53_*)QvA8k&#lFqEn1CzBe)h27e#vKYQ4PE`sWZP2qs`cW9FOs=DT` z0TD=_w!|crXS8uRzO)a69+Bi#xgQ0UOH#z#6D9=m(awk;)L)&^@A=t~I?0@Kp5F*l zQLeSWeV1j{Q&wMT!K*-rm;Evt3Jqw~117n6d40H2BKA|%9j{(nbj`H9nn`glm3mVMjEI0LZXAar zKdaGa(!0yW_Iho?Cd=EmpEH{ccDQ} zwN?GbMgn+NDNna(esTKKn(uZz>V&ar=DKM!?wDh^Z^!?B^*vIlHYviN*XH?-74+sw zW^`Iiht65%u%4Vss`|;fzhmjk->Q~dJD(zgc5@x1EBRkf?ddX)ar$>qG0=9BeUuv0AN=5^Hm`UG8CDOTCL1FA_{H0G;3>lRp%iU7!> z?JA9c>Yy1%zZ9%CX`{%wAYsfUG!W%g)a=!pjRR1HC0NR2jp{`{b!84aV)QkO5d$ex ziS*x&?&VtzW!PGGMB{G8(gw7e%CWJXEq+oXg?Hq*1?DhFW298+w(F|1;WlUCJl|o zy-7krTXehg9?+M2aTTL#?1Auett5{cshapQS}yC53e&6CB+K#&3he#TI^JzmU;BYu z82YjsBMnasotSNC62V1N18pi1DFhy0MsP*{4wH#0^AJk3490kpIxg!sDrY@3Ihvm( zs1O-YP9548#X1uf>v~O|`OntgH$fBesU`-jZ273&?jM)nU00h9uzW6(T}Dc}HfwI$ z@FF+CKe84vft)@)*<3jVzN($8Tr8wAKZKRRJNY6y>R6QXFM95CJW#avz;Y5ocM5EN z?1nM%gru8h`X)xtJJ{RB+xzL&?UcX12Z(vWEgV}xRp0)E*hKmHV`L^M{bbN@t5Pi( z&>?u-K_^4Tu5e&O_I|#B4MDHb8I>Mp&6=|9PMGft@2U2F=IS;-eF5@PH`Dn`ryd_m zVEV*&LAi5Yx$MW!N&gstm=$JCjI-dJ&dEd=o2y;~Ekqeni}fra4qU%UvwKgkp$B$> z$>(^l+LRKApJG|~Op1p~V~!!dB0|1)R$4CYtgvXe=m_q3Czp*vSe=sLLUg{-Gub@- zF$9n&-?+q+Ycr3al#!rf#Ps2Yb0|JI;yAjQzj$!wJaYh_gB6kF(o9UJp0Q#maD?9d zvr?1d#$YM;DHMYxc;@X>yB93zeiOOYqE)7M2`$XgzEUb;KD4Q1kq2X-drt(5tq>;I zn53uuV2}L)i4EzPo+k8_^M!KX`!Z zD=Rv~z%%OIuPs{eOUzv4vtv=&+!pwXF8KQ*8i7aaxNvs$Qk%(*r7JS-_MxxC;<1fB zfs1&y3R>Xwz(_bd*R3oE@H=fjw%^got*;1k#ulo}-=*Q6-JD?V zYht%A_x*Q#CV|^cBeq4=kA)%cUxd!uXGFM=22C3*?FF7Fx25PO{pfKcZ6!-qC{fyC zK@2lZw2Isq?mHHMj8V@hzuLl^)A8C=%l&>TOe_4C$#&7B6|2MzJT=@6wVqNUj<&;<>}RSm+aB}dhF@j%H+IUq zKTU<|Yx4s>iX7c5Eb^WY(Tr}INN0u!;)wzLw@sxnMB@r~=QDTz_+ir(sNuhG-4H;K z+_0Nej94{bwm0?%MqUC2UoZU9Tzb^qf&+X(eit^-A5`r|+sQh*&_sn-wR@%tOn5+E z2+$gdVtfkLUOVQq#3>VMUPmUmZ0>Kp7bJH9aujmMr%f+Wd5*ijy!8&K1?%NPl!E`3 z-#1*+HVn$kKwtNU8eptfxa40&EVL~2@AkwD%<0q)Znx$Gj z-kZE5<6tj1Py1D-yu>?#j(M>o-KAI0lNTEw7bIndtZKPNSX91PDWWtEgi4u@>pm}D z6=99kRXA}7d&N3M>ShmI#n0f6ltj2+7(w-pzfAoX3KlE44ruaUj;~!|gb%d=GRS@C zrvIdWQ{_|V4F6n zb)R>dF_wAVD0n=*xvUa~TkXB6pC=f?87q`26#c?SnIyv&bk9*kk zb3yCr_HLkm)y_*nw~E6hOeHn?EcwwZbIHFChzFWEPP}2cg-T?fROIE_#`HxReu&&# zzoxn0J#hPv*4xZmvK!Z6CX*3V+N2;OMCNH1t+Z4D2oDhHnMR~~G4oX<j%MwVcm08({mzec#cbc#gvmy z@ROG8zrYhvj#b%`$u^~kuCh5A@CJsJJX}ZA8)xi}tkbg_5>M>{6H)KDKLDVaSRLU+ ziB8>-!8&LU#qG+R2&H`cAstsT$T@I@t^Bh+o;NKdyq@ao)L=S|;f-VLQbx#}Fm*bi8hsG$k!AoxqIQX0gRR7exeQ!Q%Nk#_#}s99V=qph*Uz zsb0Gxy?rB|?0J6v@4sp`=}1R8)VBuxA*oyUsvVMDq6^F4*2F(Yj}mW|lFuQknEa$X z*)9pt30-7>5~XZKqYNLDF0DDi61Jz}aqhe7)D78}BZa$v1|TZRCft@ftoC_RN@xRA zB+ly#F1aVhxy-u|sjQY8giOVaby3m0{i^))hlQ~$(BbKknv|rJ}T6YUW;>wN74xDpzjggW=6@O9we^yUYFI zI?7@KKJAS8!2K$1>ulXV@nQ5CG|#r$C1|hl7>e-DWxO59-%sK!ih;`*sBSkkT$41e+R? zQUh(tzc%!6XcI7gwo!ZnO(O z`%z6@xaexYaD#8S6}I;9P?=k!Q+7qECVao{CS#|SBJ?U`QGCQi7OE zz-2L_9hvHNRpmhn#YzLB{M%@U^&+`fCcCXhYEbWk)#c-9gGbwT5B_ZN^N8kFBiZq$ zg;eB(HJj4#fT3vG1m=nF>cpH~)mdl-Q7I_XwUp2yqf2o~0alX%Y4D_&1Vd)Fkk##eDLUO~$LU@{lp_ zEAu$z6U47ExQ=71Z|+qpmy6~N86wHBd(a~Gu6dV)qvvG&_(?Z~TeLRA6cP6R#8k#v z5PZTK;a<72VAzBIIXnOycX$>fHj?Sti`L0gc?*4<(q?=$H0D)lsH@W#Tc|W2=%wzP zk~zUK=^9Ut)S{mqrINUx#DMEWPc*i=9^(TwBxBI0=^}g4xN9N^JnWyZ=N?X?r%mGT)*qRefPdc1M(!9~f1D&RGfIT#sM0olisX=1Ig69E<1`rTbAX#Z@@6Vu)i9(p zj(9;_xI<%L& zFMBN>d;JJN@Sg~Pnkm1$@kw=D_+tC7b}07CZH`=*T94s#U!Z&JyYt162CQ&1_)JYz z_gzbA-Elk1^w^{Uj+sKR_f1+JASOV##-rwC7v&z5e@|gMT#(HV4LLb8$rn76>YH@g zu^R*_Zm0TMgsxBYk`#wyTHj({iz6z0g1$b;=`FgKHvU&=i`vLnh7vBTksJTwjs~<{ zHm)hjAww8k{UZGsojzvm7oqmJ6?258>y8}2_9e-eE9Om?W_b;V4KA~oJ6o%Vab#}* zaoBa~RhAFU)W7_dw|+ z^4>|!%+r|dE0dO|Hz@E_D zln+S~Zm@;z$)Jo#>gya8EVa)^#;&ITXnfc=jB^W0ePxt+SKLO$JktxIfvmSRKK{U`D7)!eVMH zbY-E)_g$Iq3w42A@tyn91}v#%!Pg)AiLT#fON#mS7bA=T%DZN+3~l)68$4-pv|3&M zWOQKHnmE50MD7r23o#Mq$~q=jsu6~vPyZat$>YgZxqiXxu5--v`?Qk6=uR;K>ARL! z3^1*%C}j-qRrB?w>{sT`=U?V_s(x?Oh#j+y>d`yQb>g+7$~q-i-?U&#wtN+gI;C(COhk*xHc$0j@ZopOpG_qfzmbIU#ZWpTqKmS^&d+%#kfLUlTW%wY;ZFZa>mEsE1Bv{r$dinsBFs4uGqnBg&%!<Q+}nzaJDpCPMr`v_Y4kOCFu1$kZzw zwVivHK6@)ksbaUWlf!6nqd8^UEiIj1M<9W%!rXmPS{X-_ez=tQwv-sbYxIXw&8&G~ zvPHU3jnJnWG3(DN^@1V)e6YrJQf;$fe&R1Ilm*RpB8VJ>l?m5Cp);mvZzmK?f&Qx@gJ2j{}`)Vupt z_!!3+Ox|sA6}9f9Vjh~1J!O>3OMVjJ=u}yN$poya1Ac;v-ohP>y5+|X3=ZSnc7Bie z)7di@hE46dgN=G5O=Rj$dsNXW()-qS?dxb?VLvzLCc6U1-yxQl{^CK?AyalQ>=B_< zI%y!|<>%+WF)7x|Urq0luC_A$R=&TO)W;v(*5h5KV@B~XtGojcEtda%Zm%Qe z@Dg02Q?u!%npMyn&mO;wdfSnzk&jkVQ5~JX{DI0?-k<;X^KYqZkMD>%H3?twi^1%{ zNuzSH3U~6?WgikruxI<|YN8X}mZG}E{0w^!^xk51V;Uwm_}RWQKt0^;DVUhQl=!$P zjAr8SR1a1^h2;$54qA?-HMjakA^u7f*m$ZOOc(zrkM=|K1STUQeG_fIb%PT(2JmZF z*Lj;?u$R3JdYks~(Otwmb$X=47vCk@anG-ntYncsNgIKjOFrtnv9#m8+Mb>D_hYERut~cnCa#THNjCYrBTYH`O50hK8<*10$M5ka|oM@=vq<=Vk zk;?}7gm0LS2)H1+Q9;Q&Dq4*mKOV_FQ<8rCkmM4q_o%y6Ud_O)es9s}R%+)&uyCcw zF6ziFEsp56C;?X?LNd)3kE`90K$O0>u{6FA!wOkBnE&$GtbD;oNqEi^l3K+k9Jzi* z=Cc@JX~hV-gwd;YyO$kMk;@nR7;3z>tz8ZC=BHj6YsJ)JAsMP)Qbb~NUb}jX%v>fm?$Bhv#ob#P z-DPewN((K*%)|!4s_*DYsw{t_dL+_spp>qLM~?^_A&%c6VAe^O!oEFhcPDAvdevI) z`gcHEGGaf3`I{D3v+rnyl{xDn@O*rJq{io}Ekgt5s@fdf;;R8E2bh=?=T84 z#92MWNe{n{0bJJ$4HBc;T~_oS;e2YS?$0U6NeodT zy;Vrzj&ynB%+T4;)~}_UM=;Tb>%aW7KJ@l;m(aB3o-jp+F~Ov0cMgFT_u{*_`H`Im zt85Lb>D%XzY7WTH`x<}z^lG&NOo|4xtA8V4CU}|)`B`JlJe=lRkP81<`(_0(NE`EM z|EJ>-$T5ldq)+|1uz1VE!|Zcg@G***qGA*tUNYxsqz(3k`M3!%=Sbr!^H-)^Vm@tc zCRUHDSah3?Gwj=6 z)=iEcmM6cK8dzTvt)B97=@D&LAMGspwaDhd-6Xh&V5<&dt-cc$uFx>Rzrb}XHp zKU05r2&NGVeBWc+=`en3Uqn42ER8y?H3`7&I3~sR=vAhI*-}Ecjqk8cqK&GQO?K@65sa6|Ej^Yy!N&e zk`pHItp05AL3vRO1^R3{b_$30sw$la_O>{EbaY3lD`?OjH?NH){$_7;rH$_hW>%I& z;ea}jcjizgrU;f3TQNi>7JG|aTT7zB%yW!H?&BxYxn4inzxPSda@9GiUYiB_9nf=^ zHPnGUlx@b+!;2E<^{|{FvM(EcYrI1CvJ0H3V=F4K%)dSm9UyMbU#rXU5>0U8E_-Te z-Mj&$Hki_qlat_yKSaDLWNg=6o&NPe3J%kXA%iM*A2%fMsoOILgzDXGfldMA&s?ud zALu&08i;ked;5h?PprM}YAz-nsJXp0QOh&@^27PyZ999jRkJ!e=3u&`*F3SD-Sa8> zCq(Q|`;V@CNcd9m+-(HnnO@m9VvGz-T-nRe|8X6>XoiN;f)r@?gQ=`KHKZZ@Q>Akw zr`q%Wq-|-{>6K9(ny^XfXpcvs+25mR=K0WmG~Ju3UyJ)aNG+C|Vb|!S_OUsrDd4{F zmakmTUAYE)jWcUOe?mpMh0LS+&TemT)Xf4J5}f;bBAm7LCxiaoF&~QStm&;nB~?__Zd{uMrbmP5yOtq15khYFbamly%cx*d zQ9Uwk+kW2b-k|3`4PYvF#f=#`vtDosv4z}+xrP>}$`}|cY@Zg+zT5V=GY9E~oEUhu z-?J8aRwIxUF7L{k*m=C7X*vp;9le{ASpVqdwyHOVId{Mi{L5Kp3?(|<#xS)|&Lf%S z3zu)>fe(Q7zCTHt1>ngJ8aH&S+hE-F(?~C+z$Ch zBza~=TcToL{jMB(Tgj^CIy5~G3y`Ds5jQnf3|3T#n)WCjxEG{!mo9c-;mbbzx|KYs zLJ?k*UUCWE8H3I0voGk!pU7*!Wy@D8t^1+OLgyAD4$prW@iQCv#}7D` zRkPiDEwlS|jl#0N_C067_rydpkxwi6YU0G8K5B>U@(L={tjH|w>K#5x^HsAlx&_1h znN}&$*ylq$MYIlvKP;5F2Efzzww6Bh$kmBC(qx$3Aixj7%?CVX4WS)XLoPj_=C&sp z{6B@`T`Hrhc#~!uNM0U%vru*zfQR2|6Kjx}8qI^?1(ZN71`3ei7`I zXBS$c!`6|8b#SS<^>39LESek$<9jt0Eh1|}(TeM8z469%+aeE222}p+K-m+}IDZ3W zxp(Mk!i;G&D7j;bQ|J%kG0ajZBhdbQ#!C1iUy6`7LO}hTYWLkGw`;Cm6aw#AtS9x~ zKJX+gKr6G(Wt-Y&<~Gk$bhfj8;d7Vld+f8sC*Ik<)6u#1=TvlOywD|a_Div!7OF%J zQ8^0d$Iq}QQ>?$@2@p^JAEv&+tq%A7zfN4XZQHh8%hK>;x52h9ZNit^b{$OGPyA6O>_rY!0?e&8d-3Ox8AEn{=hOFnk9h^6q9 zP>Z&A^)8u6+kB8zu_9hPYc?~ZXphLwKTuTxi}3JmyshbEs;P*4e1QJAEidI!5mbw6 zn9N!bP0n7x_BmIDij&58U_D}$ zoDV2uU8!l$^2-K=fHp#rNos*WHn%zoOv0kZ*h+o(r0e@J{ErY+7ulU+gU^xY{hI3` zlfYm0;H=m3A!x%s>fa2A5*3<0G#T8$K!{w|aSMMZI+r!)SC)qo&BI`}>o+<3>XJVP zjq!f9>|+-g(*pd+KB-}ZCEsum3%2XNBH!D@6DAL50GNlbXg_UZ(mi+0ilnB1qigp4 z{c}Z79Y`IJF-2a=U?YmH4+hweH$l61e5qpqR?cW?Cvyp+ble3`wBXF&toY_>S#3qu z1=nDV>4f!~kSJxG-rG0k}lc*J<^@qtcqFVsqxt z8XdcnS?k$npBg~#pEHzTTYs}kfx%J_)cK|ItswVreKsWrRe94#P3n=IkV_VKMU}tnay&5<2PF6yx_9bniUsb;HNZ}m0iDx#T}mzKtme+7 zYTtXXB8z6LM>a$tM_%Zn*6uwTUwkc;4L8GGm&hydswio0c_N;FEZ^XZ{?dK7Z{?j~ zwRt5$eAzi3xI{QsaZx>D9iO)G`4J(P%Slmlwf5CYgYdrCJ15F31rD;et_mWSu54f7 z2^IzPg)#D&#U)iCzkUrEJTD`SHtKAhCjPKj35|!~6#Fv93DkjUeeiScr8j7s&Q3aA zULLz>_)C2Q$l`<{t!``WHRYLaegK@$hK&5vJBF?a-b`4;C16ieOqW(F>wjgTO%{Isz_TO!w0ycv z1thV9hRp5_&_u8-0L|>ppPX=hM{r7zGv|ZX?OwJV%*P*&wH90?cRjq%O8D3x+>8Kl5}E;x(Y{mUG?Jai1fuupv{dfRQ@5 z&w_@miIz@ZyN%HQd(VkbJ30H{6T;>Tk20U2I*7Auqr{C;*5ZT{K=(;H5q=RplFjP* z%8({)RZ^1}4;Ja-cy~}z#Yh${@MAlDr?AGh&ee<1!zscU3n)F)mo1Z>^$!4vii_kR zqLtb43l|(L@moBrGG-)ZI8V3Q7VdqOSO~w4A1F!fu!YFJ8G@S7VUFj8Dj3y`s1?A; z?~M~9vm*8)CM0QU#us!%ot`E<+3nx*VI#KS@6HyR^fbW46I*j3aM)rM=~MG;WS%rU zj?lUs3a=cExcVA&#U!_pUPzq6(d$O{&7U}hHS!{EJYN8(SdkVA_PAe&Nm*2QQBvhy z|I711rwb)RA%U#&Sg3OWJ-ak+DPm8QUVOs;z;kv;)F^5D>9EqSV=b$3`Mknv-ofu! z%vyD)$%n7wFT{%w@qyfKCLhnW;Ipt_hmoR&oPLfE^esw4r+e^%Tca zbf46^+D)h_sUuNR1B%VNc@YDfs@#5#*epA zDR?Xcp&wDPPq@({R(H$mLCWpHHfgjVG8ciLje711b!y_v8D0AwY3Q zj($F5tBh97SnymXm*FRRZl2p&M5KS59Q2O`z&3c}dUzHauz1Rzdbea__fCOouh>?< z+u&_=H)mLv18|kDzDWCf_wwmSZoYtx5 z(@S>5SLJ+CNsG11=OtnMU)wB?FOwH%`=JLEH5_W0cg=n(HLLuW6zsV#@&(;u&^@A^ z7u@nkyQ>^n9yWE^!vAo!0OPyM-YR zxA3~o-VWl5!s-OoiZZyu4C8v?UoD+Pe&w;U6pI(f%{czOL1d5u$A5)RBzS*NlIw4u zJHn+fvG)&C{lIR@J^8sXIbno%4>u~pLqd^SYdD+~=rEz3t*uV1cVmp#;!WKG#q-&p zp3j8zkQJ@yChIah76MIHh=|fyy%k{)8PFT94&Mi%f5_Fr;ibBmHefkR1Uaa9eTgFPW1gA#CQ9GQFp#B7Ob z4!cjb(Hq*-Av6V5usu;4Qs@APW?9ds&xuoKjoHt)EgiCr2sgS-6Jf=7WuB{%3DDJI z$kR66^?|sI?UueL+^>_DeSB+e(U}YG^LhlL+1pW7e^oq5k^xwTB8M%{i9|(Oa7eeh zQ4AtuhGL9p;Gv>RSSxjgNp`Le$vL=dpkMp}FMd2FXw|BDzcEgkLAt+afNtDn5#X=| z`=2{QS#vtnChoVB->yT9Kvmb}XBWp)513zXChlij)euVd?^-u1Wbb}fWE5DU){qgZ zrzJDqqlSY1z0Po84{_AWE5~aJkpOr0Q$<}My=tzSGP@si#w5<9?8O;;0<^s_kvA{% zA15)`tWiykb-G}p_bTOP9_}vxy}nBj3aBBJ8ZWX#P{DmFR%T)Ds-8y1r}ic zdJx^4IfZA?vrlCsJRm+MFRXHTy$bM%lbY_%(U6rh^?D~Kuq*&$;1wz|okvoF#5&AR zEpb+K3qxqW3+{S__M(tI%Ear_68-Qor7Wd!pT%WeHFWt1^jmO}7iZT8ndy zX>_O7%jW~Z>=%jDwKh1OyEqG_^6>-TpY+{o~YP{S|x3zt~>sWT`_$$ z{^(280(SX(Ktnxx#99$^2p;zx$JH?_^dT|S0M)M@;Tr6 z3`T1yVdi8_b7eMm!_+x$z&FI$Vk>Vz=XK=nz4t4HOnS+{VZ&OyMTDYCc-+j0t?Jq^M$1j>7gt=teov7DCa0{He|&bX-eO3V_yU zs?|)M36N8QWFqV`nzfLvq(?#_5OPS)ZHjea*ZRA$g*v&!xLfbAV-d+a<+L8CNmewhl zEQkUd)|I$j>}#fZ*hed6l5We9G05>U7TqT2-YiY{pUK3>BU}M6P3)0P<@lbkFGoH1 zcFC<#N5I$1*xznG5xIf!KfVmo3?3Fjve;)pPuwZCK2jM)5d0>|L@R{zP_udv}fHGs~oB zMi)MZRCD!Jjct&=PIeJYmKa5Oqw8MkpFwJ;p)hsfbGVgD#*`sZ0L$>V0%Y{ zl+9l{s48T$t@e%ZpF8ob1DHn3_A zo;N@wx=i<4R1m+e)9>V=_4g+>es;l$g=8Gg3feV&&=_m+R2mb=q^b;2UbwbB)MtwT zvU#cAc@leAQYFvlW6Q@;6p6CYp8s}(9J5dQv`1^R+eq^og6|hyUgAI!Th)6mhftG+ zG2=5%)h35o&@RaR-OO3xf2U0VHtUk&d_&pJ^JOv0L4I?0V7QY8JPRYqVPnZ1(AUCO zR6iuH2hsGCV-6~~)q@Z;dRSQ>kJT_WIzO)vM?v4o_CR92)#%r4``Tcmnut8zeqOw# zA{|?Y3$Me&hG-%etEYnV<1|i~6ldYS0!Z7DB>4pEBjQ;M zrBKWg(W6hXu9TVp9~emN zO9lv!I%hlmHE3bs0Oo>-ZFdC*8KL7Q89zY6;ByMq7=)7dkgsc^Z$P>a_67NL7!Phl z{rALE9#kBkpCA+Q9@fSTT?z%3C4F^IQj070$v^xRx*g%G=^~n}&a6uM_XvV7;|+*6 z_oh-p_(!LHcF25@r#rj-dtq&%=dhw4FZYRqB9V^-1&YwWCrPKmK^}7vy1@xQJ*u{J z>`LHFkRWUDWSScVm*fXrzNARa1seyU5z3i@!U>1s;_z#R??jZ_CdCJjlMT;A2upWvbrF3na%zaMYj{> z@-vA;Tat9AGkcyGSQNWgwD~uM1!)(RA1{`svxDtIP8?T_Jh%AtjVn@<)VMNJIpLx` z)peasnRXx^=&w{*nEP33-5&1*lN;y6j1A2y|Gsh_cnx}7yOMhaCW+Rr3Z8H7+K{D1 zki?RtTP&2dWqJ-tdYPm|0Rg@WM0Ml5Ab0DBRm;w}%)Rx2>HKznypfWtz1**TkhIG{ zw#nPj{nXqX5S5Vyy{p4>>iy&U@Pk#tEqw)l{;0VhgwD+zL%W@Mz2)KR! zyPz`LV0485IkXpTG9G@~g_)2#W!k4?r3BbJv;!e_3b(;Z(N z3ckE?xo|btrzIeHF0bk+ePq1zQcplZSxwZ$s+lFMWZzhhh7ye#44~N6GCKmeY1oyq z7J!^)oGukY1=rmUq|hQ0Kz~gvi{vY|Bh!K@7_FZUur{n8Rr^ z6a-4bWdC?#xWdPi#b}X4aC<7(m=2tYka?>UnO!c4A{l6fIVT6#ek#V$1gSN&6>C{T z*{qNh(b?NX9glx{Ln`zJhkCLU{1#=#txFx|hw~ibmX3veewk-f<)Lj*(xJD}WW%Kw zbwaoj;Dav_6;kU0q)ZJji-UwMNd+m0*zdBcf>S*?{#Efj*g~{GCwCTif#`s2j9%|2 z=tJJ}l^)o5{^Kd3dA$9psj*=fS%g+S1gJ01$3(mXM80Uedz%|y`s7Th2ustvb0T;x z>2s2nOzmT@yNg7g$Sa*+g!xfCHsRNQvz|5!I0^Z7#W-KRY!dK*y_fn&Uvr2RWtXd- zZcCY>edHg(jN;-v+24dmItubpl+T@N4<^`0JB={&O!;?60aW-lB z%+4s{bqfM+%t=1+q#i+DMP*pBltT#;dP9&e1A9kZL8IhQoF^0dr)NYDC?uBl3Jt}= zDw21%3KKNSAm51?>HNsDygiNjF+k+|Xt{dh1idmF>u zF&UNUeT@aylI0E|0BBxUF-;@uTt2yv23j*EASVN`eBRP7g%C`cAX9YRp1K5cO(S?$wBYe9 z;H41VI>0vk3t zw93Jhlr*RGXbr29Uz(WtaCvPe;#?)Kc+H&+gBC7g-NFScz)f}r@YZyyNw4CcuIEJXt9TVT!l66* z&4Iu>Mxj*Qs|VBk5~kgjTN?Ffe*vAykIhTo&n{BMcSIkp{~qW8VgQv%j0TT~PjeEI z1;?_(RB&|Q96wKUI%9vt{NCKF))!Qn8w>u0jY|#pjJj8a&WjBftTr}Tb5V8S1CLZJ z!S#z+z}4)wC`*~MCxMYNLgEYx9|23=Gk3hlN#Ggtms9qL$7|qz^7-y;sPB z5HHAJy|vfgOJ#=Kt$=Ll_%QHWOMtH(eBw=Ey$0_>mr7IEleInd!ND~l7rwLe=k$H} zQOQ~p7^cTKb^-NG4nS(s@?r{g%sv8t0vkq|cRNMxKR25oB1`9gvj70KDEGlRf=@UC zakuM|OQE`k!nIZ7rUvMEh69U~8Do|*J(WKSc+-W~x^Kh{@QfNt@$UscW4vyg+7HBkI+nLSQM4o!8?zU2ndS@-vkOo(y) z%_bKp{I$aEtQEyJCiq$V;%dKZ=k*lFTh(e?a0F|UJ7k4geQ}7{ehYTwZGE}B5q9?K zVqZ-ikZpWyiGMp_S`R3<^J9oP;BGI(R%tb^_>u4t{%ckKr^0 zedtWLsPH<>9vxe3oBaGZVVLr+s{whV?=Aeqs0= zM9GZ1Q;&LzR-IT=%TBgrPl*O^&M$9e=MfZumZPgOcOpI@Km|60p|3gp04?|~`+ zHiVp|Q~ui@nTYce`cl!4NtZyaX$ z>>#p}&XlRVSI!y%-@#cktK<2qXo|j>I%FMT;6v7WsZFkFz;1sgsi~GFEpiW8=CXc) z_JI|-c9^L0+BA;qvjZMja+m}BtLM#+fWl&~_}uutf}MCHg_6mrDk07cYvJ6P^fJrC zh5vE~B>S*Q8?TSKNI_#?`DzCR4Ngy#ybhe!orJ`ni_K}*IH4b)vpy_KO0yR|laN5~ zpy{w{H^fVM>>x^c5W5+h7C|Ji@83@OzV-;Ne8me<))9L-C7+xhq_W{X*6Tpoy?7BM zoII@FYx$xb<*JxF9k42(rtCUzRypII$tid`obED`v9Nt80V{jE}& z&+Aq79!vmfB5sfhs*z;QEycmft25O!{)W)}dz^sZ56?M2zwxEE+)2gN35A~XE--!? zAGFtGGoED+ka*#U#~}~fWVJvoSE!?7W`}L?hw_*t$4~H>h@iF=@Uloq*3?@p66><| za*yj+QmqyI3U|tqsQ1Bgl!_*^cF@>+)mhVJWB+K82s)~bETdrWT~5(~q;mSj%+6mJ zBLzt@g@tY0iqv0akF-f%mO{DXB#*~tx{R>z) z9~z{UP?BX`s;7MZxT?KPgB0lgH6|y)CETq5qhke&AU=CG7eb7Bn`GKLKpbct8l)6GOSc5do z!S$A{G5@e0gPvYqWad@j{f16+*lT)1{A~bHm2CI=VALe=P4TC=gNE(;RpzukhB@&^ z`TzJxx(LZE{Ld)nh_dz}Hbkto=q6j3ZOL06JFMkdw)Ux;3=~h%rdakS7?riX{UDGO zgoiYiD31`-C&D9h-K7i+9%o+|;;PrclxM{fwa>G0bZ&QYMh_IezcYx8;4A@e%bWbg zcYYC_vbhYOjl8Qm$%%kMnK@w=b${)|e5UC^9Wj6%NIXU+I`RyoY&nkEDJGqvhQ|T# zLg|VRz&Wrw9Tjj6uciC^1*(HPZI&ZpwOP2p{$3W1wl| zmb=o^nrXl}{A=5Wy$vP0+)@$gA&7Y}C=;lvCC730VJ;rMn%yv&rd^zYqVL>E=M+yS z1g*Wy6km>bFbl$oU1p5}e&Wik?WZAl)tr1#07!O#+R#(MWs3+X)d*4EgV?UTSIx0o zXBWvA_=st;hA@T648c4}LZ5LhDTIgkHs-&HcZp)RZTCj49HQ2@Pw3PtoI~oVwx9mj z(TEUfL5Z^vdAgPDxN|AiKOb@=#s~siDBZuLA}Q;#vmOi!;~+yui_U^-ARgRn=J^d0 z6z1%36Lcv(JXe@m!7mb1uQ56Qx&?PUZsZp-odWsAbXU@yn0n0#PV8=3?EFonE3erBgy|1y>@~>$sPO(=55u~y$M=T5t5E!!&J51iOSPzx6nEgKvAqS ze)Vwp03awj>{<=JDTx_N;iX)MyyoFcrE9My0Frd`_lugcX%X&CuqSk%=C5OZwEt}s z0iByjVA~%N`O`?rHgNR#FNc)@ZTclaQa+N%K&@p#Vi z#>zF)kd>3h`P5>5&Xru4uuC09fxk%E(oFfbEw%pbWyNRby^VniC5}*ffk90T@@o(f z&(6^3h3@=SKf`|#i2;q4jqEcoG!@Ql2r7Jr1kb}0#{qFs38Kq=BI`4WY`F6ze64F! z0`zVVK2CaWOxU|l4?4t=0kq{dg|R2dW@C8?P446S?WmpC^2c&&{z4eB8#4KbsVD6Q=27&{9kOku;?tPx{T5IKw zR~xwIP~(yojAiyD>pt@w$`M`55j-IlgsEO&D#m8PEc^IyP(NIeTP$-pH$Sz6F5EFf z9WSj|f8q6|0`sMA>DLz?xLkLc`xZow+LCovBq~FTfpo1QCrLv1?h^2+{1H+Q4Jb5w zjn51=MxdnCvQlyPJYO9w5oVLUM!YEuy=>m5=v<}kc*Eq`!<91q#F3(X%|8b~`yl!E zM+Jf;N<2a}sxfw>^n=o7_|jaS053RZuMDZGxrXy{*DK2I^HAZP3*m-k zhPAhXN>OuHI^`}c@>xy|HB-kSkjn4R>&k-GCrHS#gC!F6171#?T1WMK>t^~Bv!BWH z(?|tr6P{3dXX=l^)qee=Wff9T=KT9(i^lvj2@3s-cwuM$^=W4%<-}3LC+*HmF7_8S zJ_(kg8|7LW;b-u5Yr0@EiqE{A?GUAgOx=^ppc~^od2b^PG*vJN!$(2mYBLjD7xUW6 zb^&X@9gxTN^yD3czls%{;7K2ka~@$F?pb@_71#ur;pA-DjJiDs`PuBmOG-kD-kc>ac>wgY`S2^`@S0g@X1s%@vJY43PMs@lJ?@L2(gkVU;=7pdgSc=(utoLa8iy*o<{|Nxe8-L~g+# z#4p4oPh)fv0W!ki|Ne$HlsMRm{3TXZDsmyo-)jxYG55>>BZi+-j>)u01geQ4=!}WV z{B&hvl6Rz*jWKzX(_7e3CY{!VYqHjNP_GzESu}KG1H%3+WFP&=z$9G@O5R9o;?18g z>jeyIBzgsW=lv6zCc;bYjqG6b7aK%Yj$5>+EZlHDDCpv}GEc4vdCOWwRT#&bmVX4b zBhbMlx_-~B%u;#p2>AZIsIf1B64!Usjt6X7<+;lw^W1pRgKdoK54`5z;H*C|8_#HnJuDrA(WhNb8P?-=4e{1??wSxDn7Xq;7xdU2$VrJyOqj)R1HsY7S1E-o9B-}Xbyl=g3a$&p@~66qlIA)*KSwPvs0n=w{*2+ zC=!rTs-+L27z$=;kaa|XjhTUwq5ZK@4nu)ahEuAD-|AmszN2eEvIPXlrlR4x0n>p7 z95O0$i~xB*lL*a2gI8mYO$$Cd;ng?(lHdCjEk=UakWT&gw3oia8Dh7Za~IC!(XqW? z2ie4FeP%{fECrJ;>Bb_Md;+INeQfzD*Ii{Ktm`QpF=&VaSlNy!=K~KS?L)P?MXgFn zxues9GGULY#G?E}u3{AR5qQX;&_YCL+-xWuxaYuiu7#v$!j*PI1-DAQSOwPW45GQ| zccgsxCB@J`U|(D#ldh+cA-xo9@<9uS3`vZOU%&?h!O7l(Z}o{Q`)IBzLvvr63SS6k-y zPP}$U&c?P`>*&O7Z2HyTI^H@5Fjvz_L)558uJH--rP_Spy|)47VBe|J2S}kUl0-Jv zxR`gtMHSX(bRS_)5QagIxQ`Qun=CDS%+eIOkB2d;OP=}~fO9pd`(?9jeg&k^6bKQcwJLtG zYOW0rVrSQ&k>U+iNN2kFUS2k}Er$chN$?R8ko_L(=D+^j@IC%RZ_+eqdU9n1 z@Lhu$N;Vzafe+kzH7E9kJ!@al1<~sv0ze(9YHnt}o|`v?Oz8q`lWxjpdZL*0f0~gB zAc;hPvm>kb^;-gjA%IquP^EL~!hKJmf)T`!QJ4<)+ppF6!}pVb~J zrSRkBlvCrlc>Y3+Y=!6|&bMM;4U;ZuiQu<}55n4N8XIK#>q~09hy7K``C4fEQMaaZ z*A*ywyrO14O|Vk04(j05fTZAtc4tTuiv%k13CUqdJYdl2%glLTp>@6ZA9Dkft6ADt z*J?rUPEtAC4c0DMRQ?6t;bbalXLxVt**y|F&D`~Vab;4Tpl{ZPtNA5mUq}M^>NEAa z1!5f&*m#gDwe@{>WCwf+g*D+&_FtrNldL;|tN z+{wM#Lxq!k&uCA_>=j@5+{j&ntpU&LuFWcqaHpOmajb>-tAUb4>>ne+e6O3e-igWP zuhX~~8?bMEDj(Eh-fc|+wUY^C$Wt$DoNLwNcxGFxU^EQtOOlBwGm-l57nC{+*r0mp z$8Q@?1j)ztI13-ujrsj)KX~P(S4Q@|mGW;H1zpl%uvw27=YgU9$%R86M^<@|$+n<- zn2Us@45FnIQDz%*;kr_qx})b7suy@R-y))lcM6j&ajH~39p{{(GiCG$p3gIa)#As& zVpHzTg%>;)9qxn?cuh_+^d#*$aU?s44asE-p^5gIj2{?m&G?cd0iLjzW`iO7f`vB_ zru70cPROu!iAI$k;1FFteJobx$JiI`!R2)02HQrrX1Qr{@7%NXK|z&QgK!t(Qk(h* z0N_GI+S=2onE@2tS!#zUaeI$l98`%7Ca83(-;hXL#Q57qL3!+S(ExY`hK*WE_MTuX z4A$3hf_#B+SEk>hnlJ0Iw<59bI|8xPdmrVnH(x)Ky70LnC(#{Njk9w_M4|f%w#wrV zdKh>n$t}TN@Pm?M!XD?lHAnZsxx*dRSVj%-n&^)SP6AY^KUmEmjM zY}4}AIUKggCn5fwgl5jIa5DdJVvElh=nYcW?Y1Ep4Kd0Ds z)T2Eec@}i%x1RD`G#ogL?0%J80JX@&0vcAt8_uoP`;s{kniF5ry=?m6!E1#F6}K^@f}(yTW6;r=?2aaIlqHt`9dAfwiE7~ud9yy{07Gw%Zd~ivG zpf;g`=%g%LGS_gQUF5LaYA|NMWcL#x1jET2$o0j7fT_OA`z~JoZ~0lo1dcWd1EZcF zV*6T=S!3j2l0TPVUzR4vj5Ur2kbqDL_yd_n)HbU2T}iU=4SX}U*92v#&cPPlqh&81 zW#+b*;7J*cRMs@IJ(yB{ZFGV^sA=!y=$EH%Nq(bhi#wKQ##3l{%zvSEkXr~7IF1Ei zBxb*3jJKzn2;fbFgRLJK`uHw7TFM6&5XA@%+bn<6e&k`o5IlJVFUL4Zb~oK`#KejE}2B#hmH z@;5k#^KGSSv><8?YV$ADg5UW|13?LTCLeGV!cJ^K_YV`b&03Y0SCPQI$kN=0++-4K ze<@18ASq>-v+T9i>y{ZZ5^E#YuL(g|hwEZYwB^9rOS1^la2$F`X}EJfb9TK}8J)$* zjf8sQGGwMvJ;jouR9xY*QcT9veDNRxTfLf{FYNcBIB=Wrkh^C>DOZW`g8)?V(eJpi zCLR+2b=n{AT&d#KtKQvd?QQ6j%eXF)NwLmQ$ZQQ4&`eRW@%5a-Afr%GJ_+du_{d?3 zg|%(TKLzq%N2nb+NttrYM$Qg&tl)5T5O_FU-g|226;w*{kbSI*_OB9(KXa}xooe_? z9Vxcsg~oOESK56aw*YH%&(5qScozg90UC?zFA4}4L#(E)8&5_RE}UrJEpMo(it|wS z0H^Xk;vMWz1}fB^uFmV}@?Yr;ZihI{!W)&iM^&b94W|M7(gh7eMT(3`f4`Z?D{7pJ zHK&)ik^h7t(B0iC1jhawoaWVP?5vk){$jaM@h9!#Gpi6&$O44`cmb9smR=e;p+2W} z9%R)*Nuiyr%xXDO@c>jY2e$#!n6TDhYG==9pb0IqC%tSEP6)g)1*yBie1C5c@%fY!m<~hj9{uxgKRgI$ zV1w&;*>ALNP<7D14vsS_7_=w%H!UN?VIbFq(;pB~oE)tA2-)oIxpf$wu0J4MQa-;4 zJDuy_Z^r*j8sjc;B5zQ2wYF=0HoDI(TuCZm!R*UtSnb{0EQAtqtvVPNgrw0zqZz-O zjL@Hx-|o3&ur5~uWt;fC9zkLLrd{!eyKY1tbSO((iv(p)EC9df*;C>jTXXE|h*}zGJtq=7h)Ml!2J73?|_6?vC&V=)~|n=5X%@STZwLmA@)i>i1FVtPa>4^ zAYf2{HOaw94 z>0oK>Yx7SJoO>tB<8a8qOruLFqiewVLjF8xzMQg99BH?`!+-N20?5{>0r?KOqxZ4LIO5DECu;l>7ft^Iz-L89-mAEanSya{?O&^S^bv0(em4TWF|q#`1hJy~F0y#QT(*ow7)I+O2u z&cxT&K(CLW#{GZt&$>8VkB7~8@2Bs$;x2k04?bz5(2XIWbMEJ<23glLV@=0}BR?O- zf%b9(WE;9~~B|C!GfqCsSfxi>wns(5gI*cl~8hB!jvbGnB2}c}P_93uM z$AbiQflhZb2QDtB@-u|~O?lrPn_qI@ZhABPID0mz z)cP-*TnGTaXhEe6m#|83DLN9IGA}Z&Ah{!*)TDSR#}PlFgjRcQ$d2s7h-K>w5`a%; z%tYO1LBkR15fl%1;R=Q;c(kI}f~JlpXx8%U`Gjj!U6a4!dq*#5qot>MuEu$+k0*R^ zmqEwDJ{K#&SMjI6{R28<*AJ~JmU3ARYVUC~lh?1<%UAddF{m+_ey)@l12b;qc{wg5dN=O z7#|e{MsQGQqtFFcC|rElnZwFPtX=g^Qz1-Je2yfJXO>{ZJarn3LHFn0(v)`aI0Hym z)B%AEh}C@YIlVPKskuQz;_8>`JAk+INns%@1_HS5)7H1IuHUuj0i@!fgCTYsyp#V6 zZR~MzZXdH((Hvd&^^TP=aI5SWEF9&g;yUF~*l?g*sP`W_qj%*71Ua)|WgsKwzOYl= z&%4AVWc$jOe>G4oaKjf1XHAkwPS|UZ=+c#is7rZ*i2kAH!zwp9ck`G9Q4(neGx4bwBwX7I)Y_q0zrj$_mm~&_f0|$N5&NDfzq}4$I}?1 zKEBkj;ZdvRhxFZlZ6tT4mk4?$+L7Ky-#MFOfOR)1WA~g@p>*yy+r??6LE zp2;0hnA_t*e>V&Zhj^Lcyw_nCKWko!dYj?~??JUH4@Ga(0`fm*Y1?@d{4PiW)Vc9F zz4V(M&R^Nklg0n4V#}f;Vu5o-+|g>gC~-6137cdYuWiIKRPB=2jb1t4umTTI^)s=o zEGi-hn#ZMelS;W7vT@+;PuOZ*ucHGHP zMRX@gZYzQUjV0k0o$sqsVd-{G8*0K*IzZ~%^w=bUpcbTBB+zInuuQ&U6c}I60O1<$ z#&Ru?XLB(VgL%5zdv2-rt0Zqq95MG=`W-7Ye+xJ+BDgE%M|}3eQ<32~)0iCD0iq&% z-QA8)j7U8=%aN7+8TTF=9HrOF4_ZwyY&jhiml$tDneXOkNsXx?9^OM@BhhfnPoGF&?kgvGcE{loXM zen4!_*_phk-}dakXbmhopzgqf7tuzVbx6HjHXk?H6r)i1A%Q?q@eWGPAxMkMGxFl4 zK!#LA?I+f52@b?yiGFK%X16`NX6oQOksp-Vu5j4#+qn*H1S79RE&BJVzKXa3 z{4^)IDJ9vZnu(ZD283?1>Y)Q%x1sPze`<288o=#j8)uKz0l$FXQfv!0_%z_sa#5H1OS-HT?qFR*z&{@e5lXXSN zAI$ENkcA+V#onW@!I#$R-`}U^Mz0UC4NbhX8Yr&NlG7!CO=$`8YbUSpEG|{C~zt^H`+^NO(NfO4q27a+FMYwSdO&z@P6n2BPYh z9@oOCR6A9q=>7*ZPACGmA?km6Do*vmM|WU5)q=!eClay-26 zvU?pw5%MSynsJy0_V|i%bz7gDqqDB!t}&fnue|&vXEB6dH;F?g&9Z4GIoDpg7%K6` z(;a1#=|KG(z_Q`8CvMc!bdzfE&-ZisPZ={RtNbT6T zi@PCob^JpUf7TOgsm{H z+>yP#V35DPw2eS|2V;-_9ZxMKv~b?S~UhN;dRj(Yx9{n005)%1PF>*r`J8xX$d3JLtT!Oq7Y zK9a!8U#&NY>}aC$fi@{@9{Lv%gm=-Hyh>BVywfVXTE?maoaIOy+wJw>f`om)XmKCC zsR;>NGOF*lgcKD&Nl7(-7F(|&1|Z{itsw|Y_}U6$>S#L z^6FgHFK`j?^<|gxDSBTh-<1`c5Ux$AU!VogA=7bC@$@x|OHfONoV%pDUK5f4q_`zQ zpfEnh27I$kI*+Dm2`Ac!)klef#Y24{%0yV^;8K%2t_OLtfwNs1(#8qa^ndl6I*33M zVyF&n0&gb!iTmA|M<{Q{XhAWz8N0oXScx!H@lpdRHIBFh<&D7-4@O@O;fv;1p}60u z<=^B8Y6z?!PMA~rmwpnShIh&rou-32yoXppy3J7sdfD>RfPOAb6HNhIxZ`wuV*t68 z?qp`h+)p)6s>M((e%0r}-|dM^77~4?(xBjQgvdS0x!2u^Y)^?dXlRp~<6~Ty zUkD2esPpo^gcq6O9@9(Yp(4x#PbK*63VotTA?I2y69vyb%*GCWxn<7)UfUomg%T%e zgGh8e@!LP_@JwWx>)26~CW=bc}!L*$usOGI!fwG+-p0Ikl$Bo4QS z@?BtkYOh^E<9{iQ196cg(R_@{{8&Dn96rOFA!ZyvzVgSW@hF3>wb}I#yVj;3~p6ojQ zSXcO(5X<^<3mnf;UAnU=QqhQIqP*qP6Ff=Y0z&7MU_hrkZI0tsr1S z{Xz=~4X#eKWU=vcdY?8KYi81SZ|}LvJy4XG(!D76p)xU1L+_tB(he(A_68gw@v{22Ha^2-|fGlVg1ap^-X0&PJilg9#LYfn zn+^$NaPqf#QBT=@(W6d~qP}Ee0xW+K-)$%22t)ccJ=Y@mp$yC;H~B8O54+`wjwE$b zvD$&U#uKwqgMwj3=MT*DERm0fQB3zx9ckr(fLZmqQa$&-bKajqz!0?7~;P;35k6QZ7kVo|MgYNS@yYYcW#J^!Edww}`ZRYUArY zQL&sPyG2J=SqC3_tb-G&u?Y|{1F_Q^S6Lk%X@Uvac^8>J{fAV)Q=KD+V0`q4(`}hx zzyHsNX0=ae;CslzG8)Qs+VaVyQx)Oie2AxJXQd3MNcp*|A8Y}1*^dR*&OP1pO+oxk z4`|(*GC*Xs?;X=pV-nZMmKvnpfxeU5hqe~i4OmF)UG$xS5HL0jd));u8cJO0m+D-O z^OE-iEYgWH@&bHmOruC(E%ePn^7TbfEbMQ07v8hyAg)Kkmm&WyonFiA%8Y6DOgIt; zI2q;d-)6hJY<$QLSBIFIT! ze)Hz$G|U z6Yj)Y6FmdC=pwuRfWMR=1NYkAW9hZcThuav45sHtx@M$%yEKX+RhjUUO|i+>+=MvZ z2@w6$`-yjPBUcbv>XJqw;j6Y6UEexTYrg=!QN?ZH;I!L+*V*r=f1vdyYm~aR`}1|N z)P2v#1lBpM3dhQeIIov0<+=-prwbP60-i0`y8P&fSckCtl*h z?g1q(Grt>qki|lgaX2#*BWTP}8iVQ~QX!ScImxJZQkX*&inerF(;I*n+r;D z%ENMF>M>x(xEZ}Y5ok62IUh_}|L3-Zo$t7(67KjElW}{cXB{dsrYc&}s@ zr#GmlC=a1Ira_50K~n4pG5dRWP98`hAbK&@%A(^i(l}w!@u=pdxCQ)#o`01tERbzo z|Ls2PQOS~VriyV!p- zPazUkObbaIBJ&2LyGW}T6si~wZ@7S03m@h;uuai7EP=GkeDrIGWSWawRUP(n z;`Ne-_h=thaNe=`t{2vIZ%c1BrM?`IvghOVwm22hx#fxZ;7|_e{coI#5%A-mDzJ11 za>&GKiR1{01A1a<{J_`Lz;M5M-tlXL1KoK0p$byih4!)5R@7~Ul(4zN?hRhPJM;W~ zumGpgBCQ?g*fmVbuedSN9yv#apKXxqT=sw9a@Uoi!~~dOn@5?s;4K=&8DsC8)nEz( zoC4(H`qNr)uABAd&|7|R*BkBij%Ago*HXumBU-Up7f&It`5*iL>-t^7q~sgTJjB$e z+(D(C>V2lv1iRu>yqIKX^B8&9-k9^{g;KR05?G=EEMJ@YV(;INQ zZR7;AG3}u&o!J-Iy-SN|G28AbCjOBRNZZyAt+VEd_Xjq@5QRo~P_>ZC#s9`X)e!Cfu^$XfHVfQfWKnp>z&|2|-l!4H;%Wy9 zzT%!Fcm^bTK{0KKpJNU!i~jYJ2mlmg4KtqU8c_nAmFnx4@>M+`r`_^=9H*ktV=gbCVF@s!CADXHARlLoikpEY6$PdHM94j&vZ!+I~>| z$E=*mS~ljJG>cw)56mcWGnDk1bMl^>JykCT&(5Zi_lQV-J@ReaNm+Hpp+xPy(G$<3y=?FO`AI0Mo7#K^6G$EB|8xaZv$2s9f^4_vxA zNYq}eXhvN#35L`i+H)Mzr?|r6d@+SH0~v6hJ8f=R9;Xk)C?AU3lKU&JTt_-WCNPcd zysZuXmwKL$0%vb0&| zbb*D9cftR@CXF@xsR<_-b97V&idSg{8-mZGA?l$`g)0ce&xF;)D&2!(Kg>pCZD2^J=-Zeb_@MW^x$x@ZSM2f z9oaPZ0*9=DE~`8Dd;|;)^XV&xUKigV|7eZ19nG0*{Nl70=z<>lSa}*a0jc061Ig8q8mX(W*Q$IJDazj?~9^3yB zgxu*ZDP343^vqB+~VEhv+i04uIX%LpEVq&bRk*f z;097yN!R_GSAk? zI@8ZGIyn{ls!=^iZ-9ejVJ=J%8JtRF`yZlvAM9b+1DPGYG)&6uwS}XpGyRI}2!eUc zB_w*T$(RkYD_L$+t%8C8m` zxT#g8WS50lJIx)6FGk%-B!MLu+c2)gVAiV($87vnec6CB_J~Qx;uhdul2!^I+zqN@ zu4vCrkgzO=*1qYR5>UrAZ)@TJZ0`?ro4JFXXcFIeA>35|%YEE#Bacz>dIV_U%0ucL9D(T`f4=EJo)DUMP zU?imk$MAH86h27Z$GNf9_L! z1C}8@1w6JZ*xTwK+eDnF3#?jGver<9g(DC1Ta7(6h>w){*7kR-x>(i6+q*G@h1hO0 zUrwvo5LU*NK;$kWMZZTx%%9>?lk{1TkZ_T`#1KBE(`h@};*?LGsqCtYHm=z^MTv*~ zi|>02U(>ro^*tGVUhw&HB%$GAk<-FM6yH7J82wTpI2ZGh`r?=(^2YV};(ADg<44nq z;y-BurayY4RDb5l=mhQT6uXfEoGxX!%V|FkBL^a`LOD&=i?SQ{S220k8Ok-P45P%i zP_DffA7X{-?$Xs&rH2!%ex{I;G!Ua1oxyL7GyX|$7v1&q61XJ(B|q|tQ*%Ru^w=)y zl?L71-#7vHBOx_eS;$7>*B@^VeEyspEvfVo=T!(z6LrE>{eDMXSu`51|Nc-bfX!g? zx?pk;^Dk^)jbRAitE!@iI-)B9yuvb7g-|vvQ{gMw){d>fCv2i?DjL4dv^#skYYW6M zztTJ0A_mTYD4AY%q*@VX9rPM>0stw~x330< zjDJIX+8hi^Xgnn4GoCC=#4#^h$J>Gk-HOH@6O#wbW~T%dWB5ISM-PTeKdrE>uDpov zas?<=ZkG|li{kgZ`IuSkMRaO}BaiU!!a`DO99=>glDb;8UUPUt$?A$C-SBvX$Nl(` zctwEcA}!kDVZP}{I zBTiV|y!FXrI`*bjJVt^`QDj(IS)KU#@tdR1e=*mfZ@I{X!6r;Aqrz{aD>u;`lI)TB z+MG()7LqjWPhQc?g4hPUaZeTg9XPPmf?q@UT}tvJHk?)pU;9ucv;BFuET};xb>oXH zK}FXU;WV{WDGn?9-C9?fZUI}Vs` zSF_^@H}Tcc9S0gC*JzfSBS6HRW~1sV@EO9yPYtnlJFeagnf9w*3S~c5v5kA$dQFC- zER$U!o5G#4j>wVnQ;lG8H&0hItm%K^>)(G@*Y67v%amkT@UqeRPu0WZhaPEPW{X4% zcnHpgYBpgYo7XPu0Mio zSBcAY2#!}Ub`xiH(gXigfZQO2#f)s2)@;1L_g-b7oY$BZVt2V-`VZJf!eoZ8c>9-~ zX7uL^UXCFo)*K`m^O5p80&4dN*ZBJ6I$@@0? zcEv+AblNFCO$+RbHXuY(*l9#wx&w+th9Qg6$XcJj{&h?>kEt;Bfu+c&1y4UnD9`mb z+LbC4+H7!>6N%FNNazn%E^LZYp_%0|C(f}egX5rQwEO87u!>bUxOk;Z?aYHL#FT_^ zq#Gq3{O_m){3;{dM-EmpMl^n02`JeyAzxdcQ2JyeVnXg17;T%+am6Po@q2LJNxJj@ zACg8W5y)#lOU~A<-QmBLAMQPlAAaDv9lc`U(_s49y`gb*_n&-fnjeA$i+!@!lpM~Q5>fF(7?j?_7-?X;!uNf7a7E~qOv=aS`Z1FwOY!MF@1gMk^Es0 zDTPZ1S#DBRnB`IFrLDAnddmZ5ynSP4;o+XMq4(OQoVPNMKGHW_KIC=;@8A zPh-NR`E6aQH=r_WbC+x~!qau1qFJW|8(pJuNUogtNg+d717pRS`hVvqfGFwsZCe$c z!RX8Gvmg?~H7?Bf&GS5hTi%{!v@ElQ0d823Kt0xL2>LobhB5aJuC*A^~KxCSxKV~_8e>*>s< zi=;nqbvMYG%qtk!?XQL6piCZ3}gr5}4f=gyyx8Li^#OJFMH)k`#5ku8Wz89apwQ!m)R1PYp* z3Cww31{SJ^c7a!$2ch0LjBJZ7Y?$4>w(i0Wz`ez3$R`Eq@@5364#u&V1N(4vWqO=J zBYOGjX!UGoxE;Jy=E}{Fxgj>x`kxr9Uv3p1WfhzNrvpm96MCLV7n!r#QWCsGZ$CA4 z4=6Tr3fw?A)8W5k)^p-ZRCz7CokHgFV zn=~kvuXU1}^fyT*`{e9>`?T%S!0}G$Yy1X6wt-!=5V5htoA#x+bJqtNuu}yUJT4_X zE=pL?EF>|P?_!jdsQYt(h^s!S_5S{VNeul>?4SXFQh#xWCUazc*lDwpf zcBk35NW)A{JU0IuIbTe{T!U`&Jf&hMzK6kv$Hgo(K=<}F)qno*(JvC4ONl^XOAwA<#L3M145;ayf){zmPgk#Wr6g6-_Y|+?GujxK%RcVR% zGFS-K3beVpLwIIZ`dWXSzp1QW~k#ZyOdVd@X8Xo3{a_MCEWxj2W<4X-w71|Au$(faXLX*bjRmgdoTjp3O zoJ^loN!}kh{nc1Ja3a|3r$Efv=%jTzWxvAC?I7_!#%y3G0sC&37OW7C#u{2_Tqti-NirC^{{<3mRu>A? zL6S)A&`{&yP~*jqu%{y5=WM#*)|GZ#-FT!b$IKJ&{2BF*+eHYFWm%;TKER%msV(Hn zuzi6&ABNb%Bp7j|6yt7NhIvWUziq!Ps$%Ol(7J3q)akXK9X2`PP_6G5COYA;KO2bb7_Aml_d`(TJ_v zT{EF!UD9{Vq@87tR;`^5Rk9dkkPZ}Y#hBldET5CRXjF)O9WK(?{4mc46T<#rSd4A) zic`U3{j2eCv23l=6%h(jpT6h8#V+}GPlqRsh(uS;q@d2E+OAMaG?6xXS`ktmH z1z4wxYZS8fLDZ@9;cW^v zX3xptbA>_T1A0+>OkeIoyaDvS2f_GcCb+ktx2plc&*kV3f^w1i8yIR%AO5eVZE>dh zuLQSfpSJ}cTcUR0Jv)-UCt1OhZ-JjPuBV5o0`IpZ@_+O~vR*k?>EY|Jht@Iq*^@eT zA{ROf8ja87WH&nnxH##JLrC=Ifo&To*YTe^c7JdN-KxsO?k#wf;FL>*yfPpJ#|6I8 z_p^oASaW(GxFWTvam!44cDXuwK9_4=-n2$mZvFR^ zRZmEojIk=KZHrXQ>%en@&v>TQCrzK1rhfh=t$ zI3h{wFZOFE zck&qt0~@ud&ATaDW<)xJzqBGy^v-zW+yO~4iRuQp9OME+ZgM}@K|J~*`18Zmr2;-` ztRHzf6sH`V&TKuyncJ@`lvXf3h z#+qxh4mN!mO&)MSiae33PRyzSrtvjy*ML{q1se0_4dB4_3_kQ>9!?qKSE9@{d7T>J ze!0+v^W{wIZ_jP+7{9__HGZFaTNz0_WEwC%Mn={+GkL>w(qwyQ4s=O!8a+c}+-@Lk z9b?>1D`^F@HS~lE_!X&EM#W%TzjdqXvB3j}&u~l}VM-jl@hGO3F&Ww%`0TE|LOpXS z#G%%s*4<5Z1ES3{8R9s^Pk4uKgUq4)@Fb!>PY8YZL!30gX`jZ`uIA(p-`L8S%EOZ}~T2~dP_+oe$g zr#I~0^R>`L{8VtnXJ`b@mCuVl^t92{q57`A({R?=MzGd<$Zj^+nxU;(tCQ#4u6-wS zn}o?u2l~yKT$6P%od=jNi~no}%X}4e2&wE*^m);3|1$-O50bNmU+afEEi#Egk?%5p z-Y9GQSV((+-0{7p`|${|WJ2vCvAr)yAxA5@WL9{nn69UWVhN0=0EYs0Nc`HJW-~=n<=R0(7MGDwh(z+s;{B0YT|X3Nl1-rss?gV)%Cs zBvB|Gd5sK5_XN%%(2bd$M)ic2jNzotDf%vZZvLGN@(Fy%!|`sfS0?E4G&tvba(s^7 z1=?)leATJMD^E(Z`Woh7vbZeO*S(OnfzV}qRu^gfAmLWCN&}mUn?mlU*!e3*v7)r6 zs!XwD>xvDc#VWzu`8%gzn}P?i$iRkX@{ln58{400|D-4$Z*+@8ER@QJYV`)KbvH#% z0`{8!;ZSwN>)gu}$Z1|eVBp-8Mv)tD^`YFI0b{p31IiMI+(~2}$v91CH=4(!bC~B^C5yuSyJuSV+YeV`8r;`_ zvzSLt%}L$@EKwwx@f2a>*&96 zkM;$z#e)P*cy-Y|MWpXB2)FqHi67yd2SZ;-9LxBEvn=kmP^DnN4;Kdfkj0YS4-!cz zLFt3KN*8{KGiOMg$(6hI9;XXkv%pB$2r(YGh#;!uesC}0aD4o7e@yT=cQ!rgbH{>< za~@(g3g|kVxic08yUd!C{KGuucPqWJq6#H1z~yS(gkgh3gi6TW1 zL-z-$yhzC-56^Ln9Ke{yIg5xy0kNAEX@g7|z$Mfnsg8rjHC& z2twpS0YxFjT5WN(s;0>^)ZN>)KNyNmbP?G!h*fBbcL;4DtM3l_<&;Gl#V_UjbIULW$5%72Dfza-H;S&1oN8~lC!`iy0W2yOa9O$@N_U-s( zkM+@0qcP!^6t>6U29x9shSsw?=~>G&KfM4q*Lxm6hG3E}n*?T*@T&j*Jq%*0*9UWM zCc|G&pg%~f!W`z@Nn0Dj*Ta%<;umn^NMEMFSuO)NuA0xCy}(BD&;Fnh{!#DEiL2`F zfu-9A?^BV|FvM(B))m{YpvF{Nc|V}Z04fy=@!2tAv8Z^Pk>)(3R4jJ~b(Mu=VZEi~ z)|bRBnH^^bsAAqS7OomV1!|br2zOSgA{t5i!7jE~=Z**kpV=su#RHd(+_)#w;#ZRb zWiH~#jr;A)Is>|C+qsm2F(A5W=t`a*$zF9n^-boV*f$XAIIUO-m9zu2VCmU|pVh<% zD5?FqLAo-VNlScq66o8oHRqx!$sU8*Eb272vD9ijb#6XfHuFI;gcml_Dav{J_79bl z`;*ZHBRiI!A$vb?DAE6-$0zUdLpFH2DikrV!m+(GH=h&zS^8{JEs<> zkUCj6%k4Q9;dEC20-wpM+CG&Bc8fT^T~`}j)C3CH@Yc`WBso!Bz5nF}`w|Z!v)PRD zJxPe5yPe@UI)A~E`!5FHV$QthAM{0Nc~50qW?qz9iL-6Qz(#zLT69==#(15dXRiZI zMiCm5KoNe8Bc9-)Q+^4@cML&f+tIyI%6g#Yw^YY3Dj7vn70j zRCg;p<^{5ZoohSu2EpGJ+fqu?jO*G-h)Y7w%bC~n@o^)*2Km0&A;MKv4qYTK#exD~-2N!&Bn3*ttRj(98ye+ya-xy6h& z*Cn&6cKIX=w`r4FWbUUqbz@hk-!Imm4nR~Wi(0;z{k+ms>b>J!z0K~T&`*Q3lQ?!( z1mvzizSd8Rj)0t5*56H`tT}Q)GS2G_4@8uIvES-%bfSA-qqXN|y3?!aBfY8#B5W>+ zCRIJTT0k2Z7m0CN!@8`pWR&yOt@K zedfTD4BpYxi#TLSVt3qfM|fPlk?mt`b}MHHwxB)Vv)3zsCnfgkT+`)J4*i{;y5~I8 z#nn~D0e^$SOfw+msG;?H#OVm$Vsg(@`so_$TViCg&)%(Q%bz=c9=ldKz&d6AdeuI( z5xN`i-JMHi?3H|nyA+7uvV~!~j0yGhpi*V9r*ZkooRFo?XOvvwS9chEGlHtgcHPCmE!w>Z44HyW+>b0_ z^3haFZ;twjQmPW(I=9~p#D4?k=_a_I>kA zNRaHy$l|cP<$fa$cYQ6NC{CG=_h|+P?$}lhE+>fpyQ}r)i>)+?|Ho))ozujFznv$% z_JK^-2jaK-tGvG~0FhzA5n!xB6I{ldbXl+lLL*2a&M7$e5{>rVL^6uzWLO$Tf46Zn z#ID|7^u;{t=JsJt{LMY727*{%#`T9OwLL9roMuMDEG|S7TEksA1TI`b_r0uBj7Odgs(prgFnp~JTz>;VU!V7^K>J6A5R>FKYxn8w`J!m zVz7{qr|F~Bm*wEmh)0}_L#QuL=L^3|@Z~Pz}7CZpvM>&3@=?$6S_6 zRVA0cfe)4hyuh&9pwhD!0*9{5rRm{@4(*@33FNcKK0~Rb=(i!F4XH5JhymbUj37Lu z1e@>&TUq1Ujh{3Q{O&LiY7<|E66mGZ(cnn0VJb>>df@5`JYV_3uA%rMS;{Q$eZi87 zsR%}rg5*R$#MMHW_+5x;Xj^`!UBRZV1SrY((eFw_*m+b?z=wwri@p~|TfK$xd_N@|~z@v=mGx=47>yZDiII|XC1>%NWaD{3Q+g1yq?1XhqUjW+G3nEWr zw)PyhdwjKvA*I{hC<1@sfzMG4S22^Ca9(5T6tt`+(cQ%9&t&v2bf-YQC;aNf;em%R z88HBIklXs(D|>PMaT4*mYf_=mQhtvhVXzy>+wuGaFJ9CkBehD#ZFc`YQnnjUFI+QI zh?i`I8H!dm+9}aeZ&WCp5)V>UajjUjE2q(ES!P}RqN&A#$$EG)N9}P4gca!F^F0`R z&Et1@yPves`NvIH@)smYp5$^SvMI_MKsb7Hee(Mva>&eR=-i*k7w%iDKfy?o%r9?=mnTAXwEy8 zJxw4VmQRl*mcN9-TaMR@Vr--Y&GLS_$%fNGOLj(W5efNNj;0(yyQ=SFT!<25fVj#z z*6HEDw_xqWF*@`{mLx^oUzjoar;j074IpQ*9)zvSlA5|eE5i?=5 zum1?_3K{H?ciHnyBCUeMqhnT~9iVdb7}V$nMJFuW^1W({5+v6u-S8rKlqKX zIw=5*(l>D^){otDQpw`zV}u~PwMQDmf@B83;LJ7Iee|c@U!y?<<4?Q=PVFKrwm^ml zxsTA}WEt_}_a5DbS+)lRp@>hYTJbrLPw!xo+9Wm%gISBBdG^^M^2B*QqE2-B9%F7Q zv?j!qv92Os-y2I>zQSie@Li?yhBvVPJ}zv|0weQ6v?Wp05sLF3)t3 z*KvOZpK@7E-x`$?{3K%hBOSAE4h zXm3d&e=H3sXg9LVW7KGV9ZJWM2)2qmg^p~EhSSzh-WTNdZ_kDXy?op`J9zP*)hg%s z2q(A*<4B$zB{#bF7D8a2!qO9L8NcGyK#jXwr$d{o-u^|@D?u;02{nKbD=mBvPKuW>(~ z3?C0dhV9N~Erx2f4rbfewkD1>*82Xcj&@fq%7ZycvVP>sVrM=e#*0}Ov-Q9Fl|{&} zL}yh2lE@>+m}GP2DahF{vC&rtwQI5SG2^D?{SW4%X1$sc#$LSMjy=Yr4lcT1>){w_ zpVSv$w4D2n&OMM1^^zTocby>e^g&m;X+6TdpU3n#0+lYmH@VbxSQN@K*j+pQa*$Xs zHgM+|w3Or`WuPrW;10E`V50gOxJKk1#Ij3LHBfw!%s~L}h1t+7kM&*%&lXp;XV^^* zL!b9O3SASTJG=p5{hBmQmhD%~D4}3CaCAeD3b<7eyWCpzUxnuY>nP* zoe7$t8aJeo6*s#st{`8tP^B2wPZ+J4Yk%+x1dzY3u8WQHpNA6_gJ?^gEu&WhQzmv^ zs-lFw*iUf{dd^o z5#KZEco=oHT_aJYR!l>I;#Z@+LsZ)Hc?|`B2CQ7cgai6n0|bwLMqcB>fiE&-F|wE) zxY_FV9S_5v^5Urc!2Al)#7I(+Jlyo2cL;qn?jcv4*3q1=#o@t~{W+r*&_Z)YMTaoPS}$EQfWEoZ3jftY3=eACo)QtDnVAB!Y`! zs|+t6y~0Kzd7_$pu&mCAkCEKY8RTdC$0EDs1mZQWXM@s3(zDKccnIuOe(oI3uO&Kc z(3GC;`l$j;gg=<+J|L~y(a{W#f)j!9Jb!1t^Qql}xkQ(fE*jVFOZE@Y$+tO9N*K4- z8FY(>F!Z9(lP8Y4?9*rR%K`Ts|DWJ~v~6nLq1eMjmu+CeGW12A?ne_xhAK644po3x zJFRdvk)NUZI*N539lEn(j5yEFDqoc@{{EWMv71IZXk)2XTdCRJ-~+`E6G!`Nq_M*S z8hxyfFQ+a-&nRR`5NI|=Pc?5)nDVm>_4Yw;;-}r+EO?+vhExm@>|3A(?+Ve%J*u@X z2uVMm&tE;=gdzKVY<(G&$|et!;{6sYC$@})ntf09{qcJa0z~FSxv+$o3)$N}#?2wq zc#|cgJ))0gzIAVI-WCOnVF}>~>?Y}=`5Gv)s-g<$-GhfUf#NzXfu>6$Fp+hg)>)B2 zxL(x!EUT1ag(2+VqK$&5nf!izzZ<~|1Ius{OBMdpFxeZrb;O;&q)tPle@mG$jcf$T zh-fyjsty`%rUP^wjv!UR`6Ot9!{5?;@^1&7nLDt2jf5$ zs#8{fEfBr9B=!Bh(&0307&%+Nu-RB~tUn3t9Q$xLYYoU7OR^5hwDboTbPJ8&96iOB zYQ5Dah?aVIgZtD2O!b1V#CKyU$weeV1GG-5%l8!hQ#Gcs!qzUygUL4lBD*Od0idAs zc56_n5X$`K@S<1Q)0#d(^iTh;kN3GS_w*|O6mTVl`oVpKCJE8j*}K%6{-7$tet+`K z*yC(Oms%ZH{@*aAh|gw+fZ_J@c+rj09ij?ZzqVIA)8Dwz7h~gz^t&k#^ACcue&o`c zC|-{R3e4dBF2IAIlc;7V^}J9bu0P<|6XBh)Yi+ZNOb9raY+`qA1@)lOBo_KDS}LS_ z%6T#msac9He>yPB-(KvvHacihf8}P*J5Btew&z<2@$>ApXXX)eM!(Ik{y_#2i*VhLT$ND|WHyxd$_1Ze4>tRd9Mu*)Yc$Jn%BN zTwv>1r5hjF6B-j&aBqaia9WAAtN!K(kQWTKhy@9Wx{DPJ#)^CSz`j;S5Nl;YY)8ES zp{5dmoDz0ZV)>z*NM2$e-145!ctsx9siYoZIMN=<@#J~l6P|vC3cx3|O!B7aWBx2B zIpUuu)sIb&Du{36Y`aKCd|$AZ#8g_) ztxA(Ylm`fOgLz9z{92bHBzHGtcMB68H3$xvXgBEqf~eQUL?aJMoo+!Ds&{rR?=b*J zHr_W`uCe+M!#@R|=#;yOum%-*c2BR8Vzrucj)>LO%;$AFj^49Ovt_>JYaS>UwjzgZ z+UW7lh;Tglfa9;kYf2p}DTSQb!_8w^P|AL-P`v)y<%@e?yMD}vg}NGIvI0*c5s&Kx z7%cfEaVdxm345>@F2OCs zgpHSaRyRRk9QpxK7D_kSIpQGJ>^uRS)QY_|cji2xN%N?T@W zK4NE~^J`wmlIcae%BIA%lN(W&l-2E#ReHJr#8uH~n&+VGw6NmCVHKF-&qBM|eqi?17e>JtG;6m~mQ*E@HLZViXY=e~F0pmZph z>FTuDK_OJ6MYf@%Jd+K`DGqvY=4J2S|1fSjg-=zS|K5c|(oKpDyPST}IAEfIj4eYb z3viJ$DRUF_Y}~Estzhc3!mPm7&ah*es9V2_M3%Ey%6T8*CDZKS)4f+6g_0vPR&gTv zqVBP;2hGPY>h~k1=iKbH96$M#LhZT5_p9;IMoWj;!m{3$k9XYkV$}ddO2pMsFAnCI z%KjI()&tLXaM$^Ig|ub^f(x?fED@*^2+Y*@GuR+{?^^KIogigVy+O7-F_ON-Tq1=a zC4q^Lj;es-BY!TQC0Q@Fg0!w)`=w{P`XM8I>zgkmHE9hmmBe?87oT8#QUExQ2)d@=J5-(t+({6B5a-HVTE8SqBZ_-`%l4Kp%HGMa1cV zV1ZBp{#|rp#nOJiCH~vDQT;PQi@vjugkx?g`pdUYTw1${Cin9x?Ml1S4?|0_WM^9pgI9pKY0+BkN_D#Q5k@LO1hGrOf{pY zJ->e;&7>J^hcQZKePibl6K~x3Xo_J!RW^IORpHtM4&r$bRvD2K$n`Vdt?rdc?S-9f z;ML|caPwVViabj8$wh;ZyamX!Z|q@JM&B_p>e_^lc^Rw`d(GRr==oyr*|NLXBpV)xVzI{>hoXzYG3#T<5HdyEvvi@09tB)q zQo94hCV$TScmw0G=!0}c5kNGq9q`U{&p1I0FjJl`ga-a*1vk_rw%2RVRhvoV;?~60M zu7+9iDKE%-WgWHj~0h{LNsuksY2 zU5+>;(BZyyW&jHCm%WJmrI6e1&ZV0O5fft}~~uA(takG50i9wM1glb<8RVlW7* zw=VZB0Se0b_V{s-3y$y)nq2*HnhW#v+GufHX#Z)i@Lm|04$G3E8{4Sz+}M>8zI6|9 z$*BYvuZoZbR&|Y|<{F7iEnQY>B;DPyCS75(>DlJg=3vwE{(Q)Y7Rm(flk?#DVg%ev z_LcS3OOz28tMez1T~y4ciu6!$IKt(Jdl~O=VSRc5vmMG`u7u_P zx{~@+AK-@#O$>JgGzidmHm>=;mc6hO%yo2euJU0nUg%MPIcRlS9#%0Po7m+<=*JDs z^YUF4>9wt@)Bl;jU_0)}4H#~sT>qelBgc9vUmFU-JH>Ub2rcjK@yK6#e7wX=L>=!k zzK10#59^ctRaO>Wb`&dXvwq2N-vC717W!^?uXlU)?p`Rj#ahRqpEiLu9kFuM1XNq# z;O=x!40IB2M+UaixL@2E#0(WCUY>|=(}Q9^#JQ@(h+DeOD_@=L|fo}Z^B{D&wD~4kTuygjU(Tgi`I`&hZqdaH!JBH!Tjp{z|H}A8s zx0|f2D-0RUF{_ao(9SL6_q+x;^4kHgSwExFF~Uh0SFCjyk~srcfgMSRTy2x4PBXT+ zRm#pV&5xUsaM1dU`4Q=({J0A_!>=auNm-v08#RZ$d@y?blwrBPrU7oAwNWB%5{5~& z2`e)lzQ0~2$mh`ivBZ`lh7X0`e4Y<^s}tl0rchlA+DT2`U*($}ZvKXlOno?j?77ki zWC3X?&@$tx`TT)Mtb|L%@(m+#+WV#l`}YU03fxwrl}0jb9*i>{9m&OhC|j$Zn7k?> zKY*Dx5jI7DX+d>_snnAbo%xn2C_2)hBxWE&9Pp9cp~aGv1~=}=yH!u9rdPa#yerxM ztD40RAF6mn>v@+A72C=q!4S8_)7M}-5vL-tuXsLye3$Szxu?sO4}Xs_$=6dQ+Gl7! zWb3HWXmSV;rn`eiG?7mk=1O1b>a(fF*nzqkX>SJQc&NMAj71~;9BwtaVt4Oql++$K zSxE#${BUI&{UigQS%m?N*g#`b8HcQ>-GurJkUvw2N3x*$DP|MtXD1YZKPJfKNV#Ee z#+XP%Y4E@h8~n&jwi#BYhF%K3<-SRrWdL)yEYa38ZqLu+!iwX47cxD*X{Ma})NN4O z)Rr(D%#{0bt+y&@?t}U_S7)bV@R6$%_w}VY5zF<@n|ysBpvnAUzqGMlfh#B3n^xDV}Y7f@=Yg zZ7n?mQ+KO8GmKh!rE2I{b0EtOP?!P$UDg)&c$*oDO0;nll_C6Ve%@aW9|bpm3W(-_ zVi7=n8RDzNZz=3oWQ5Y8BwQU&pYN`)fKyS7ZK-so(#opgQBDEtyn2!Qxg#9BgHFm_8&eM#aPec0>g<>6xCuYYxm3Q(fYd_fF!FlL zpnM~{@;8+EINNji64*&cUowdYiA8xQ9)Wmiir>1a5jn8p}Hk@z0=GG(8?o-c9#rI+i~uc1+a zAV+7W0nUN0xBxCS<}?oGZsVWPk_kJe~G`YA%9rouk^a&Aq8T6 zB%@%@58pT0qn5n?3aN~f-b{zQP*DAKm2z|ewoxo=Gb@z_Bv#FO$i2Gp&(f`VcxJ`X z9}?5&C4S{SsmrwbFkklq2X-8HdN`@a{zaMC04G_x8YxaQM2343{r5P&`U+PTgd#z$ zU+U7v90n-$%@<0gt64u7XFaMnjq7@iS~hddSrY7Jt{sgrV{e2T=Gl9H$eA zwH0zx6{8t8junm;xGBVz*D|ZHHD7A~h7UQaa^n^|VmT~fe$oCB>`JdE@X;DnVa4zT_rro}`R#n)XKuQcv&_i$ z1eUxwWc4f+;#L)(rj$RwnWWLh7L7w)Ql?eOt5+rTIJfhMDg5!w2V8)F8C0z3wzuGa zE6rfGJ`Atq(?prTtD>RFUpG+{$zT4Z@{kkMH7~yTdX~fw`LE%J#~^&vdAaF;94#%B zXz81(T*2|v9UdCZ%p8JXfl(K^HC;gPiJ&I#vx6dHkn!azzEUlXJ(N9b2Ca;^%p$MU zWW;sP&wKl3)r~~DAzTZTpc!=TW4%9Qth3|uoaBaek?6uo%wY_6Fc5MY8?}1t(&k)h z$5{3&`9dE@-G_G~6+5QRSMCwilAx0bJ%zdL{wP~zH?lK>0numF1|?YE#{9JXH;SP% ziWFxt+7PlZHCfsyOv?ugd1Vee45qrgV~IqhziS*|?5N_=9{wbF$Ar_(0GAdM?XRS) zC6_nRy(DQGwYqfiN{Ws`TEXSphNk1EUvE6%6awAXPeVXL8>2*I4-^2)2kZz@gfb)P z)_qcl5HlH;eavV(%zSpdV1KV7&m3XX5pw~j0>$N}Mv+MpW5pr37rj$n~`Toxq} zU@<_NoTL)mt=>%gUdgvb0sG zPxjEdKW^Hgp2lZM z-q;~x<5YQM)7;U|vyd}^f_9*T;7MlxgJlN7b%PeZVoPIFc^Z=Fz^@1 z=Pe%!OBlX3ci@wX*{99`fmCMB!#1emcXM7QT$+y-8rx~9Z-khJ;3T#@hFey7*e=a>y_Q1fo3KOm zqwwXlN5WR|5~M2p_$gnMqRbZS-;k%g78qLnCXHQMXxe=b-tN|CYgs`k(AmSe5BM|^ z=tIIx)jRP*UdaS4>a61cV(fDNBcZB5Z)tjNk=H zVWQc_98d%#5~cFzWQ1CRAb2RB4d@7rm+fl?F|$C^4t8->I2h+mXwV1zLLL&!q6U7)C&;AejB~_q-ITRiv{RB zGuqMhBh>eq@e(1t3V(aag6!sQY+p~daCihjAIo5ftO<%wx-_u%OpDZZgq$+h*Ne8R# zt8$LSc;;={^mw<_$bHM=&vY8+2)E{oCqwVLB`<>~_Tq2HZGqLPFM1M5LCF$u1lK32 z1oqiaOq=K2-MOZu@dzw{{u$XcDoE;m+#!QXwn>GanHS>RfI&mF_d5@ApS?0Zq93wk zOb=#Us_~S78q)~tj*|q2{{KNyF4g=ZGKUOMD5lkf)*od?fW!bsd zU1SK0eJzQEQ{8AL5%NX3>lUl*j=}iRV;8PL>k28>dNBsv@T}))e^43&H z_JzoNm+9blLwJ5XDDjbqSJAbho|IUFfPv6oYa)ah6v`5Iu23P%ti#+sx{CexQGug@ zwt^l@4Yz(XKurMJ)W_$`t!$D}V96brP2@}rtqqkiJ+6o)NEMoXS1$hBZXt_6>6dy9 z!F294L?|p#;g^h5n~X54CAjP96ijMqC7v^1@R$5b6ZHa8Tx|wvShDTiPH4Z#_ z+MBqAY_O8~sAR0%c>U4U;T>MdcFrSY?)#VzOuRj9QtTMHPr|)qF+isa$W3}y)K$kE{BYd&4S4Uys$OcKtXoMk(l?z4gk|cjJGYYPJ$Bc zO7JoF!c~k_FMUy6&HL>RKcHS9=tE{XRT4p=NsXmg01^{07X zu0ie3)aIIW044&KYm~NR7wO=@()g$hwyavbghI&L%?4qhFn%VDQem>#^c9o;O19y( zxAWfM;VmJaV4yR3CBuCncV~WjjP9)Xc>t-ZN!DJ6k9be&qpx1TKrzE~MAl@=UuOdY zx|S>`v;InX*L*2g@ejD7g@RGD3PC{Kijv%P<0)k2(pe&9Xn`-HfLDUIDpRsk9HVn! z<3sGCi4elWUCV&g6O;_^@l#kMdE`BLT;gc5*t>wK&vrboslyr!TbEs72@nDHW^MUv zEG!0M=vFaubD);*s}oQmsrA1K?+Gt9Na3-EuzC8_Z;o5b!1br8i7n}fx`-1IgHCg3 z&gm;yipcU5WmS|Cs#15WOqJ^yfyASQ`XATxQ$JLQ2%wy0-&O1Corxi>wVB3tfWiM` zIT~bOeW9oAY_O*x9>vW&1c;v;>U)m0@v}j=X+K6NeB?G}n?>sht()ue>J353SHw<) zbAsX-{bcipV3EXU-d>Y9?r`3E@gxdni(X1IDvN-Eb(L9!S|$-!L6+7ElVkp#{y|=2 zeN-UG$0bdsE)nNXcFPmBO5r3syB#@>nnSx*ica=rq%0v|X?UUd>&W=uoc0!Ns*$O- zgkt~+aM=C9NYgmaUN+}sWG;3a1Q@5If9z7kv$&>#Ju)n;ni`Do01XNNT${gnJIVLg zjsw~UKp5v1(f!X}ad4~f`tY87_hNtoZNZdrWR2oLC&`=h>v{tyOchStAu?Tu5oJ{i z^Yg>iOR}oX0S3z7o&o>vnh9XSh%U5#8EO2SxjC)&qHRE$js=S0fvLz+0Ha;(!EawqsF1ShU>p>vN`rercr}$_z5TRld*@V&K2qbRl}M!M%Zr4OaZPSC zj-Bt5O{MStSROc*C4aE!fH2KMN}m2UjU!ij^lzv38;1rbA?d!q_x{nb5%ar67RJq9ZI| z37iCrff3r(gOtePM#$AG_}9#-CBwp&!N3h+kNqQ^4`|L0?Y17}zPncwQ~n9HK@wid z1A{mxKt_x34m{}*y9Xe#B4WC4POr|&Peig71OTQk0%qrE%sPcD$Ta_6GnRRi#3s7b zy7qVRDFgf!sM#JluX$~@o?N;6rMOV26)Y%6PoItk63lW{s1H~$`YM|43^sK@%H*ZsXcme3IgefA+0S_e;!@WbZ zJpSy}8+6ao;`z5~tOP-GGKk2k;Y3pNJB)*sv3Y6~$hsG3OCvV3urb-g87UqDx^$ZE z$$12o#l*RNR%C$JJYs%N-t*%g%qmQy-}%pB0pRk^J~(2rUvY06Wo2OXUv~R9-BBY( zsoPcHj+di9nzt$r93nuN#M^b=uMFNU9VcABIhIEXWbm{h!B=d3sm(Ez{bl-Un{k-~ zvQL^OpBT!Wq3@wDLg;>*&0#@rD`<(NlL(CfvFNO=BY zSq0IE@OX81MH~*{LYJOhFT~I|pD$|-MCI2tT(CMz5>v<6Sm-Ru{sORl4soLOLULd0 zU^dl4|AMK&f_Pe z)=*t~IYWlrA%r{XM|FLOsvC^jWU!}iH)g1g{QXEUls>+aq4a7C^L?EpN_F?G)<%Mr zdD<2Kd#AMLR9FmH$d#If2ZSYtKTmS?^{~jMaagoEC)e-({$d8;ZlBx8t0bq`Pt5ak zlq``VW77SHt#Q6eyl?FHXYaLii2x}LULdl?NJ`w$|LRF@s51PokOtYF(A^<2%x-qN zo^k-K$>Yx|{9c++1esrV2q;DMvmq05C%+le{z#I}YpHL)K^N48>ELz%xVjxpIpqyG z{EE4CYCK(aejwVzA(_h8UgNjQ;bix44v{AnCj<)U7#LGD^OHGIi%4unw=yGK?(YEg ztbe`?3F{LWO#hvD|6amygBcgE5tA?18fqvuRbC<{|3lDA!!-@i<4+ZQq}KHGaFSg) zn^iY1-YU^QiQ#7{XYnxn?H$hQ7#8mvvbF)dFaL6$e)PzRFXE~Iod9)-3LR_Vw7K2& z$eOP-;d$JS2wlII!3z0Mswwk4Qkf?UXUq%ITsmkCq5D-wulkIb7j8k4KAldf!pgTMt-MsC#P%q0o6Dhi0LWXKG6NP3MR2g2^Fm;&h>WNO0nY?YPof;kCE0+D=6&O3RD;MMuFsx!YL zV><*o0)%r^%(9=x$oJ^I&$-}3?6WI`suuUJe`-YBJ5Z=v#K9ERM3ro7^2&C{tyCa!C;s zM)bWzn{FKUUD4Og(0{MiV+%2dRq(pu9_h#N_xzUczPtdaWNfawmPTbNTg&&_$KQXLdHDI1N$@CgD z^-`lHJ+QeNb?-d;=D!bo$s&=XF}sb<`GIf1fjD+ zoPa!{!HRx2pGXgD&(+7J#11_k{`5|2Nq0{JQiH_r^3zAHc$vX+;vl}!M|(uzM1UO^ zsG_VL^wt1wJD@%~D%03T;h1^7x40dClxIMN=oS*Xg&O^qFUaOy(>Q{xQ>Tv67$qd& zeAQ|tZq=axR*hZAA_H7^t3Bx7M)LPhktcBX<{gCOxqUhDR!@1wxqtFi41n3PI8c0V zbn4sQ#$6wV%;&BlSsD)e0t$b&d?c)#2nv*g=u|YKyDz)`6LUik|F;zOcKeb0)}=*jP3FX=jmtMy6&6`P8#S0pWD-=cCDz5CE8cyV zmsS=AF7~`OYx3w_rU4jz`|-?Ugx$FuS)cp^8vZ(`v{G0(*=s2@zD*-R70&(?Z>B!L z5~{#*hiBQqOk{8)E2e7SQPGDd5PcrdMRI%P#`Xu%B6=n13*5i3YHZ8~>OO$RQV5?8 z{8J!XxSWi#fBWEigV5?)LYMBA|7x|V(6VI!!m$XRafQK#^CUSH?(UEXFxdS2Zb27uZ`_*R@6sX612tDL~HBo@m<< zFEf^zBzX6Pe|y*iFh85uo6bA`wWmXuQ27<-;54}0*V_1%6`~r+S`P)AGaBTPNIM|{ z9s|r`xM&_>Ome2Q$XmObrg7-}LrTwq&Q`1gPxhiVXbCetdc2+Q##n4# zxXtIs=enEP-!eIJk((%~ZY-1sSb%8cBD2}MEi5~((fMy!Oz-&R@F&hmQkE;p>&&Kv zT?spf=I3X882U2$^;CZs2F`IQW%GVrVi?iu5oioom$Tlu{YA-g)ZoRz@7`G_Na6rUh=b%-xT zsy-V!X{_A*L&C6HtU-OXU=Jy!8|=4p_M+Ph$2^bahZimbr9s1fU_`E%f2mqt(m>I9 zok2hzWh5E(NH0NQw4_j98Q#iDYw^UmhD=(Za8Kif(}uji+L=$bNVb6!edcbO!M^qeSe zgIqwsYpeR^7h>SRxS}?WKl3Ha3rb|}s`Sy7&YtN0**it(_~AFo&V31osp|Oc-FqlR z?H(BugC^dDKe*9lNKEx&-U1pC9e)vFkEFAKwCW$CnC||Rd~crE;M#~w8p)1kqCo;# zAfc`{Px2e|M5ley2~zpA18T2nIV1Q+Kg(an%*iT78m za`}$D?Et<6_|el)D^lE)ffDk zWn2g92%|IuG(F>I07Od0J zctJ$=r&^)jqfDWqhy>hX-M^ZfV{k>Pa%g_lNoG3?bhKJwF4?}XqLFZ~AY7RatH_<2 zC2T-9Jnl;Ai}L15WHAVzGyFU9}Ui_;P4=N36n_rH!#E;Lr{kOI+e5Ew{F3 z-;_4MCPF}QOI&lUU6@M;HojS}06{Pt(2w$%)BetqW(ZMZ##O#eKlCvCGSOVbcQ;(# zX;@9p!ho73lCqN%JsOR%PD5NS9XKN|NaaKp6J~A|Qtn@phhZSb zh8YMW@!vT-u!P62{z&$&pb4edJLou;{jYH!&_UnsGH}kaLI&PcBNr7JPcQ@BE3j&{ z_Ko+fYM;Y?*}b7U2f9?AFM#`QVRdt7PXx-Nq#;7gSl@Pb(y;r;P^Cm2^j1m#sA^Ms&UYW1(3u>&vvjf!3mejA3< zSD*P3dk9nK+$7Z`_5_lMCicTl)MGlHv@O3#Um!8?F&uJhOeRrFN?d0L=Y=-g$8Z-?<~C z8k?tKs)$OJF&%^? zv1%+d)0h-ups!uU=byP&Kz;On4eoo?%?ZxlHkm4e%(XBS&VEdv?U|FG9t+#}@oiSV zWdi>W&D}WT^17!RNP3quJN$L1I$QH^ZA?i>eB1`n z#kPG)=W+jO(gybzRKi2Xq_Ue7Nk4MXHJh~0qHK&Oc>l{{0{O*5#jc!O^1^U=pTO}_ zM+q;?S%h~^%hOrKcYG`&frC*lBaSj3Ba{kGo4BxOD=D*j2Tt#)?P7IInsFw(aK=A~ zAfwlzFTU)*Di={<3PArMvZAY@PuHo#NIcxD+wq`Z9OxIH+V1?H_WSSm;a_0Fha2_F zFX(LzI7H?Sb^RIMIm5Baw5tZ{{u_CZqsE1VT2f;77FN_!xE})zZ^t1L1t`|du|x9w z^oD}u;#PSgiXJllV$f$2NVZKS)qJZw@GoqxUK|3;cSzJEAN}V{f}&zaIZ#~kxs@L8 z$A=`GIJ|vmDt@0HZ#Vvx<%JpJ&l@v(hXWegWU_c+Il+bfU?K*z6`bs8oJ47m*2laC zp}CN*0<%5zlyRhERRzW|gI97Mq@0)QpCnzkjEilf$O7_+4C{afiT{(a__Hs;_lPEI zw3w0^8ZO3BN-0<9D}9l_LG`mcig3qY{?|q@q=lP3;6JgGLw5DJ_H;XX0My|(Had{B z?p7lGNm6lbdPKY)H5qr}CPAfWz<6IitAAYE=4)=yTc^GME`g<;a0I1V%?PIyh&CR- z9hTLo$3tdhpzD$I$4qz1C412aK^No<*enW)gW7){QvBI%)K9SdHOd6_*(1QV9;|Wr zr}58=e^5O~ahoXFMzJys?X+yns-sDrAZ~1uWb3Iy88hUWC^I@cUgh7Wn-ekczEbM6 zPigM8$Jc)!-OMCb4YWm1N*0k({I7xi+2mHwQdVUF_nsy(dc}oaH;=i|%KInY1xzjM zn{QF?_3pXwnk(%o@b_}hVO`9n%ry@Y>h*ZSF6n)wC48Q#GLne)vzZlosn zL4X`T!MbKgf1%T=(#dA33sU4Nvo-~;MG&j5I$IurMXeUD`pLjdzXKJo$0w(v(ajEj2M_XWUR( z$l&{n#zwTt!M?9h4Qk$hm2}%O6&KZQmbxqu%jX2O-11+={rw@0ofNv!UdPdl+MH&OzbD z^@NU;IN`WCM%VLApdz1YPSh7=Ro_mwzsuckV zXYdsF0RtnQ2A5)#JcXKrSvBduGJVK|ok1`rA>*uuadTJN9>A)EN5Y18VIIYkB0kU6 z6fAHValSSNqzJ70srFhcFx(gOP*6@WyG_Z*a)(FXfUS5c4rAUDqAuI(j0Uegul2c^ zeS#UJi!{skQ;D7g9ra8T>1hkQm}a_&*Ek^y@#1;!LWw-?V}27|!6mbcjMLV@SAt$^ zLWl?|TO;PPGfMmlMe*^I9|1G_QAs{vI~n3anKR?FK_qPHu8OnS+KzQ=*~R#O+JUo zj>Ns*Iod#yr5snccd-dt45Y8w3l(|cF_L%ykwy7b&W|tL$pl(^Kl>{23G~C zYk{6NnC|KSfu05hXD8xKV2kzXbgP|5vC*ODIy^4nexkQ!8q-C(vp}l`OQH)W9)~Lx zNa4`+``t7#utSDxSBVz!(-J$J{&|NeX=sMge2#=JiSM9jlqrTYn^+Vw5YN9_sg*e*r)FIh_LDGMMOxVx` z^E*lpodOgN8OIm^=siKj^LwPyfkpUv?J4}EK)Us73T+(C|n5iNXuDE=?|hZYU@_LAUB7P*QOej zhc{14ste*W^=_xDLUP`9hHz8#u>XAQp890lNfb9(AExq8=J;>qfp`O#hKsOXOcjQD0%($q zNYLY6%`e4E_E}@QjWcXtNBp>VL(?__6lB5^hI$P+oLqYt9J6ojB}l9S?IK*Jh+@9a zY`IYmHWdN!;s#-fwBq5IxAQCb$XZ!L5mefQA;br-;VPo>C^6iyRF=A4TpwsJ6^_2m zc|3JIqpdx@&f95~7R6tAPJ$ZapGm?S{_p*ChF4AuOQ zu$-qg?%w+h6)@QSB`Yuxgn{16J8^Uo1)EU>YEYzNfZRoRxPFjtLPiEEF}Xv%BvpN9 zF29q(6P`N3+-Nis|Fd(*$M-QkdQCkchL5DR+;k}LO6omH9hA7QknJ&(t0pD^&ewMW z3sK<+nVIi-O3~$37i>Tyn;tg9{~^A4MciOG=9iS&5qDtw8PJiWl$1d@LBB=g&&>d) zBD#0`f|w87E*j%UNOE3r=@x!c57EuAHk7!0rj#;xzzRK>;{PXG@*hr$ffU*{3|#7q z&{Qx)`s%UK>_$p64sS>fMGs*^-_Xq1e5-ox*ND~vm&YwGIMTs}7j0?Uu3VwCZR8e` zN$4hSf6ED!5_>;Q`JznkjdoISQ$|w>q=?P2 zr%n&Ya)w4)j=BjJ-rVS>FGkSd*dfH-|JSF6AT)}qfw_3u4nqZi@b_gvO1VW99tN{x zQuCHQ_}jkI_R}_EmtQ#eliy6ps4qVmR}}LdMoXItSDolYgEi9coDKv1zZs_E=#PjZ zL?T8tvn`N?4KYsYgggsHygE$Gwbw^AOBPo3seIF!6FU0#v-x6NhVx#})kwke44iQ5 z1%+@HN40Zb;zJ^{vP?#Wb%XR?5M=vVdwQ4n7^PNZcHCAN28GRol>DBVgitbW&M*>Q zvT}zqqn}Asu2j43C7)3Pp@9`>HSW86E|WnG-|IP`Q0V7i`Ro6O$&@7l)o}=D80v** zw(KxtWr!j(FJn=iFrRmQsOD&r)_WqiMQe~y)xDGLn?F!xonF8HWL%rv^H2GU7f!a> z;$;OHrN@+>qH>>>0#sZbDnvQkQLOrEOUscPKeKgHge34p`>uNcxP@{@?CQ`8P)WEp zHfQyb!cu6i1;}UgJV9}Ip+N7(=zS!^M0A|HHqqPXwNP7bxqDp&ZyICHGN}v@>@%lOeng^2jnXTyb*p+w+P~?E`R5_+Ha_SW;X6*vC6{l*RVD&e2knku6I>15 zv!&fhaFgj!J4gO20Qs8;`1=j*DOPXHBPFGq*G``pLK#rUW%)zo8xurCj^*HS0sZD- z)`=vBdirdEyXVrnU+6rg3rOAz*s7*-or&9$qJX}wJ@co5N}1S;?BN?XG?3W$ zSZeJKtW}_KamSO_Ly0t z7Yhv~fFS7mGai}8fgipT6S1+d`1#69-w(}TNm)5ablUS4;hPf^sTc9MJ)L#T&acbQ z_aSoKANlH>78ieKgT2R|oD#&=%n|9@D6+*+b=e{?oP^tOgC=$(EvtpgJ{fR*c_BC4 zh9P}c8bt#tzm2r3k{B?g>B5TbP&sWZdP~(N=GK(wcqCvH3Ui34XDS&exa473MGjhVAndH{!z+ zjq*gUY8gT`Lt;lyP;vm02^li#eK68Ly+E%?v@K%pLdAB9uyK)5d7r1;m=0%;T*vUP zqW2`JPWTj5G(PpAo#UAm3B(3Z`9SP_=sZ90D$WgxmuB{-P3RsATeZb?&hHn6>)Mh| zb!cv(k`#tE#j$?7=_{kxi7qweOrclZb-oghuJ zlNR1K{<0vRp*}Rz#H)alpXWpLsk^o|3eBg7H0e*>&~B^Xy?yUL@O>=0u*as$dnIJa z-*`i+oifaq-e|)j2M+1+d^cAg=A1bA-19i0vq|!;!CBVHWarP=*?oiatEc3B=AO>v z3u!aEG73+rsORGMcAr%FR`XYh;@x?5{H$VR9^%G9Py-M+?(!%>= zgL_}~grY(U3+~}U-`+pyY0xOyi#nHk#rEJWv1io)%2e;dh2qD@n6pmL}O4*yo!Dn9}l*xC2mD$S{bs$2hYs^tg_Jx z%hRfWXE{ZG>jg#YgObKbL>k^!uE?LyU3gz`gbmz?OZXI4^qNYcsuk39^vC{MVG9ei zV9x}7i0^8i2r%I5c4?2d{D#b;dsU=5=<(;IIadPqS06y#5~hnHlHnu zi4HJAFMBf38`XWJj9kjESwy6g&O0tOfAxPL`#9|yFSoo4F+R?F1{qAY0hPzd$KXVZ z$EjVj$GcbkA8-M^^LvS_=H0N)%#b(aJ)^4b3WybDBm-Dw&esa>u{AnFt0nk6g{OEqmN-6tgJnN3ac28J z9WP|mG6guk-d_4t<%*!zwwBX|?{225*iwS;+^hg}A5%F0()SQulXo&Dqrh> z{7HG4W_ba>_F7c+JwrZvM$!u_=;&-Wd32V*yjq;Ay44%m+fw^$%h$ZoQdV=)`?cNE z>&se@fqD)LDS_sF+Zcgv^ALH-E0a2xoPlNxhvdae5|_$%eUb^84tS(Uo1jNG)1KH) zToyn2W2ZR!qGw5Uwn${*>TjBewEr&$hMe}GsPIuI|AD z!S}3VIN1QSfAcmrIEm{f&H$`*qnnw`JjI8T$WmKomp?9*l-tnOf9)Nwk?m>LyfEIF ztn-HBJljTFBrP`zqd~AB{gZMsC1>J!+}#f@?3qu#y)SQN9?U^H95&Z{eIO1m`8FG| z1C!_*A^LB!t8?lf2!N8+Q2VU@4^9biRGs3;^(dJ-Q`I|yT05i{6{8>IpX5`lqwahG z2cYS9QEL5S$|)?zxxdHc$TrQ;BUuI^h?Ui6rFMJao=*{-*hL7wu2hR_YAv&WZ?W3g zE`Q1)@}=IR-g$14Js1z&{9}9sEPJ^7&HkanZCHDMSv;(~3hpsgwNW#%O*;8_r2EIt_d1^x>T}&G0^}I;+@-M}c&|80ELXjLEO(K4 zIDA@>$BxRHa}07Zy_)3fHDpT^9XuMC>O0u8-I4arUYS36m)Wj=U%_R2tiSBUC1}r_ z68an}uwAGhnfEKPd(cxCXlePQM z8f$!M&nE{y7=M^9tN-|M=R{%WX;~Ns1&(2{DYjC?Q++A zu}otb!~5-ie$R3I-sAYWpXYu3Idk~Vb)DCFo#*-aoS*CZ=IFM}@bI5lVimS$$#V=W zXB4i`U-?oMUb#m}j6IaTYX0mjV26iu^8a9moYufv=Xm*!Z*zQGv{cZIWgA&t*_Yj2 z9>|gCWFy4+t&?Sn7f+oy)uRLeGvP@`^_}a%)Qz73*!qVW$4p#P9@>xxGPbiw)o~ZA>p3n?>rB6-g^O%*=XAn$0=SPnW9#V6X@0Y?N_=McWxxxY(2%)9^fh?nmeB;wS5 zowMNO$Y(*Hc=oCaAQfL@uF~d;9zc%Zv*7vLuD2FALd!&N;V)a3qnD&koVlj2)XVHN z*YdnG)g@zNebtgs&A1JIxGlByK+EMkhpNF`qTPNrh%4&VZZKZj%Q)w3*P_LE6<*r^ zYFU+tZuc3!V$8;7WJ&OL(u=jusaS`DljjM&%2tDr<_t+y*F~fG^wZO};13Jm$Bp%=`#)M< znQt0E8;)=yoxy*(3VM!w!hW-RwJdifMhQf!XuV_ERaHozG&d=x+(dld$(ruM-@=N# z>bbJeP*dn>K-+SEB<(q{rPXJ1eM*jZb++r6_&a=peY>iRj2$)=JIJvyqEIa23k8#J->3mmyqnpEvgsVlgv%Vu(6deebZZB478 zb24&iHJhd)d|5LhyySG}E|K+GKCDd3KFqVFYA{&1`$UayVfF1TE2W%$AEYgGpf@g& zWG>xcTVy>(H0EOMJmF{U7}4w`hT}Eh3PQgcGWcTsQjUjBNI=qFRcrjrFBSgg+G4I3 zJ0X{Ln_q7~uHIx4utVPwvDpoVw{sj_#5PR$e1F($#nH~K)yYpp9}%m+)uiEJf}>T{ zPboq>{HH@J9b-cCH>E4KUtWiOKu4QDef>7`R;~60i3CMHr19nOGg|*A?)Z&GL~=l1 z#y_{@q_UKi=zT73O0TUhhwB&qpcEWI1|7XY!wk-pPFLYbcXk$^O|Jzd)v3Qzj2WZ! zd6<+QJ1(@-l|roHzg_sV>o94ktxLb=Mc1j2jbp%r^`5jg<>-$z2_TBj&%@fO4t0Ht z!GuFh%-RN+@gZk2JRwMvv6{Ubz4zUj{BY!A`ilyA%0nil#=~_xzE-+N`mGb6bo%1d z$^9FT8sb=EXVfn6&`aFn7vM>QO!~!DNgjvL;e#*(u>Ev6q-)G$wOO5K&l=PgZle?P zHF7$fy2qFqHC!;bX=sjkz#%U&`@27F+&`anFFvcUW`5JCg7h#>$|^k9%DZs#j2d+L zOIS%1KbY6gI^=nb#g5-IR2J6DY%?hBtdCHmnkW$Z)b&9&6OdcGD)etuEBC7B4caS# z-wTH-}_@_`YOKTFrruMm-Fly8G17-W{(ZptnCd#TrwN1&I0&o5$c* z#V-EJEqju|9zHRk91*EDLa0Fl_qm1;zoAbru7+*)<}qK5-ZwHJKV*-D zNh(Na9=xtZw(M$uiM`C5XByi?5f^mS<*OoHo0tH#ypmDI_9keu61H(twPdQ zha)|^TTh%m$D`W2@#ARSGK^t;yEhL+ni;lsDP4y*F*_GB;MM|XCs))qbZi|nj`gh8 zxWx|A%5Lj$><<~FpSz?IV$y$!$$pB%Yuq{G)?V@#o7z<6ID=F^{lrqQfjwot7N%I@ zOYZRzRM28wanSAD!6TP=%h0auv-z#+UE%!|U%J7Oan}sABI}CsX3tS{CHW8c#UES4KljclhN_XK=`R{dsBNw~hbr+?s65YO$}6&Q|A*eW6LY*PS<@B9+?vqQNQy0^!kw zG_@WBSGb)MMGW&u81M<+-q|>H_2Vwwf>05<>fjiZI=*u1Mt1Lu)FUw8-7^SK=1#ki zt9%P+&Pv@>7E(+b-%3&bTGz>XL`MBt91AOkXKR`dl0{FLy~46xQ)Q#ymp#YH;o$Of z)GiLjSGO=Cv~bZr_{kJ|6#1lIn-R3N`g*CmJ^2pUS+&~V|9RJ^D6Sgq8l&wGfWIm# zO8i3{BC^@hJ`;he>E)6N15S>miiY}U7-cJoVUtRKfa?se^z86P*|=@vKHLr#+o6s)0zgULjmj{8H~vow{_#F5~B>-c_m7 zo0sUiEJ#-5+D6`^BYOKKk_&5lG>EN-{-dDRAcH1mYG}+zV52z?V#BW6Zb8k^_Zdj1 zulB2*7`tA`n1Ha7L(y&MXRy&0*<(C`_fA=18vLkQQs_7f*Upin}MDDV2ax2=yB>&>0L zpQ9Kyh`LfVEEhJBNUQ#fV^c@~A*{2Zdv{HZ5YO8YekMlr_5+V*76em%BwGqf8zwjoGH`G2;JvW~aP zjC?fs$)S6~tzaC!{2Fa=epOOt`Sq#)i>>)^;m7nI`R z*U8#^FYX!G-*4k%{7Q9UTGKulx!tpt>XnRWNm;)9<;TYJ=qvPFE(+9m`#h+OK|tKB z&C+vo9oz003@w*vBMTYBWur-X6N&#H}9N1X2`->rl` z$l}&})Ve;z`*IMn^(iaUR^OCizkuD+R%l6S1TpHU`@ZEDmmHD0N--pQ zwCxbhN)2JEm^20AZSfiK*Ke#*swqA-x?#jBTWwW) z7t_3BlOOBw7_}1lRCOFwSsksn;}Wu*aa8>&CABygdg2}9wtFEwRLWiXoqM6-ebWZr zrr=v(7MU3})22+(Zu6B2JUpP`bBtpIw|cANH2xbe$NH=N(OxupK2wM=XWlg9Df1@3 zhsB+Cj{0(Gj2Ml3F5}XgxNk^9o3%2Vq4PZI!k*2zzVzplGQ&b+N)|o5K70k^?oBML zIfSmhPVWDkh6N1Lp@Wq5{w9j`|Gr-uJ}V9WGyHnMHgviZr1}jeUh$KYkb{1j|Ayi) zg;fhq^{KjB__$eWL1q_EKhzDrJND+}kWmB)Hh=3$Os(NDp$Naff__zuj;zkx? zUVTw8$X|hTS1;aAQgU|`K^nWq9JM+qoBXB=Fa0^j%Pdo|fKUE#JnJcT-{F}Z-2ts} z?;QWB3wNGX*5D?eFz-1mJn*InGm~l!*?)QJx8D$pT8$5Vr*&9}v!b;KBF$se=lCcqA-FbXAgBeh6~l zqN|G-fc9#FV_dilv{SSxy~cT&q$Iml{|4Km?*l~{qt6`g?wlm+$W1zPpA9`Hioc?| zbWN|!d6(W2vRnu@H5W(DB0CB7e*O>z7k-~ zbv0mY%93iTJ*Djz1t`nzYS%N}cxJ*-wE# z=bNNd6u!jtZEU7ys}99@HG;T=w~}L6yc8h@+iPMb4dQ$4giIymLPFR#*ARG+Z|&5W zIYcLEwkw3QA(>9EU?(2)003xze4EKX@H|F`3@k-0_k5nI{OGxr(hO;GtDp@kyqKqw zbQWXmoP}+HGN1!{;oND|>9&*in;*_s+y;mNir2dT*->#XAtM4pd z-zDKqq01k!PdQU~ZKns0%H%2ZcIw8yE8j->6ZmC!sG;jZge7|>&<(u4Zo#G30{tc2 z(vTZuue^Ub!W431DU{yMXYg3z7A3+Y^ji8DnFUde)r}p@y4EYly>&|BRKvjq)dTfA ztox(KLqg34D3=F>JeOaDQ#vH110TkJL=;&z0#o&La1eI zaN0CEE|Cyhe#CX~$yInh91*dGAeW>+vb1YTL7k{RnUJ`|M*XNq2cIgibJ7HBDR~N$ zuYh&Ni*$aJMa99ri8HF5pu3=&^u9EsDP^~`Jv-io)ijj-E@qWO*A80aZdTVLM{eIH zWYhWAJ#Q^5N0vmTk5o`;3$<+*!jo!S4C0Ivg^qeRB&2Cd_m?EZpUcPCzv;}PTb4yR z>fp>RNEHFw3nZ_pkeEpMN?j*}muC`|-tiUaAGkC_OE?Zr0MdMZis2T-k3))qT*dRK z^F36@rIzSgROq;zChyBXx~=YC_6zo1cj#bdHn-=oWR zNofKcxB=$!d@R8ywB6zJ8#(EY&tjHue&)6A!c88L*6hc#iFzY8U($JGJPo(mNQjUO z7dw1MyY6S)b1m1G&!{EKC?8~gsJQrB^YMmb`^+d^j4{azVDo6R6{) zCUqqj%kHg{Q!Di0GtGC}=cL>oHH;_fcVlnHJHF{nBnb0P@JzyNJbHVA zX{m(G7VSQ(x~M@YA#+|2c=2|Yi}l-g#D~PZh^x5>(-sOT6;&p=)7UL7aLQWsSx8r{ z3fhLX(Hmhabc69+stAA5az(}d+6!*u5Q1dl>$j9>GM2iFOD#Sglbuwk*)ck$Kz(a`Fg z_~n9z7Ykh&{9|y0KYxIW?5!yi-Jn!Ezf+}dhxkVC^j-O&b;O+JCi+XcY+)vM^QBWc zVn(@fBbGXX!O$k$O=1HGf5Yv#B%7AWmv$Qu_Iu_#R8kviI!RP`>Fk|5i2?VukHY7- zK22eGH+h>vED0-fA2EyQHL-7$xXrxg9CV}wt`^ZjGYx2!RtazfFK?~jK71A)Guv7H z@qrQ<9Wrb{h<9RwqU&_VtB|3~vJ4e`QpRC8GjQyvJ^H4N3rF1ug#~OA7?Ycg%b|UJ zpMmdEjQQ@-k$0BX z-&k>~mDu)oQ_vZZhk3If7ZVKPifKweW z;sXk*zn-?bXRSLglY8o3e&bCbVRk{Q-M9y(-se}$5hmn3fBNKD+brkDR<`mnpq zU;;j>HCs{!H3&;HqMIrq3JAy7!ef?$!wMm(O!bB_p}`52@6G7&Ii+rO9#TlN>8sj{ zjS{pQljV*r+ywKFjY}LN!z`GPmx`x{l zo|$v2j)f0nSw8m^>-2S8V`cwz5@>dzk2@W_x+2q}cS9MpbQO!gREEi>-41^Fs7QaY zrGQe-6{)Oh-7RXbB=0C0>0+^?MHr>HqWY5AToS`t6@PXtS^p74N|8rlim> zS5z9K$|SOo#(%V=_C|??oDdQFwCjlTq6)6opd8MfWJrlSK(3YxBXS4{4(oBWPlbf9 zB8kgB$S2BaKF6Q;Iv_QLUg&(iWY^M7n#XSa3{)nElY5RjAYZv`Z2U5LG)}5~B(szW z=GgRv3d||c8IX)AQ1m>83;Feg-<=jHw)V@u(8?cXV?t{{>+470ec6La?{&LY6U9yp zqQcB5;FtCAXl@1mL3$j{W-O@%%pk0Ng&&gg{rSP}t|5p)#ulXE9y9PRP1W*JA`;nG zRf8Hq&~xYdvX|vT$_5t?m4Cc}@bxn7CQ?T>RxQ!|WEuBIO$86fkMO~CuWkr)cICe) zYb|iqHJERiYZ+c8B{G4x;Kgok+(3ZDWm8`LN^II|@oEe~jWW_)XwqPvD1Wa`*HPSC zT6PqvgJu@+(e65SGM6wvAX@JHD9RNT+!C|_96Izof0nmmVPr>GBG`9EaJzpmGC*u) z_mxxR%sKL3ywSO|^T|o9E2foMDj}7RhhofR5NrypIS;+h98ANp9Kt^+xJ=*rB2>(( z68%zw5*0IDL4D9L=GeV=8HX4=h8V1vhh}QE^)1qOTOPwT{V;I58()or^NLRA0SVYE z3HDQyi*NbVlB!n=Hz~j=E7QO?Yra8CDPFRlK>A~5x#;%sqgD!M|95MCww0hbgPCMt<6oSI^qWapbuzz)Q;dhKU;%KY} ze!lRX=NxyW)|zfMwww}wm~a1j=wDzra`4L_;+;;(7E7$Tzgaz7(XTX%dbJ}T#?|Oe z<}$`DI3Ho?NVU!E|8e8*IRV_W$}(azI;`|bA}5!E%1 znZi#ifLim+VdH-?&e6IwHtKlplS`*?o>l^i2K67YB6Z73p`mGuW?TYalM8}y~cN<$JjhAcg zv*~o1FZ4RK%J6ZF)OFd*Hm;@4-z(dsHfUzZKHAiy zcBs)MzDqvJK7LkhD;a z@+FTAl;pjIgX0VEUdGTR)1VEA+Nmbv3$;g*((6Zc-Ll{sV2|DzMzNfz+jN6sh$aFX zv~~!c*nXClZ4BA+??<_m(gajKLg-&speK?HT|sYr#E_0SMB@|&W>%v-;80x zHm$dA2Vze6D#Pm6{2=jXs#A^kQ>Uy>^xdVlc0ls}2&K^_uri|JE95wDGDLT$sj6N^`PQj#n=Y<2GlaK8=sU0+7k1S1-9E|GWRK; ztyIBVO$B>R#zi*gTg+55UBpC$Hyk$Bc>Aw^Vif?7zxu^ng1FU&-kcw%NBvp!IBE{K zn;#DAOXsBxUeoC}ZBvrPwY(LY^DSgO9-VL2s{JVkp5(1PZ3_FY$bYb5GNkX`(RlT; z6D<8}+)mvv?9JoM)%uXPgh)*GjXsIGc{#}UUp^NalWs6{>b`#3xcI&XBveQ=*jE%KV-$4nMrxID zFZ*tTMUl@~!aLg_Yuv(`-Nf;o$fRbIaooLyXZ99_*jm9s4|8;Lqn$_Pt$ragHO4D1 zXe?8^Joly|nTH`bS?7;?N zpG|5FYZwd-*bC^39%J=rPHO!6uSgEbYSk_`Fpd&XY(I@Y`r~bd;fZ()t9vQPkbzK_ z+34@KiM45QSrU}m;4t5{?2!iRfca-WVBhpwm|wnFy5jrHNs;JCGuqe%9Ik3Tghx8; zO$e%EFUV_hA3rEJJ+qexKR*`32r7|c^lcLzEr!=DsKIq!m!J?BHWQ*<+ar?#hDsOJ z)TqZ6l)!G!C}~qV!_uzPM!2qlJ5qfxr7P35vEu z*BE*kSlW{wn~09Q{u#Hz^gxw4Je#fyVpG zOq0utUeDzO8q^kudw9zt^r0o1yb*AK^v3V@vxH`Pw$c4o;z@AWczKTek}(y} zj3`r+P9t}sQuEH=PHPqE?j8Acn(FYqO_lU*Vkvyz_5H^I(vruB)((wudtdM)CR1&7 z#k7Be6ic|=!%=%>X}($vow(Y&H7q|T1&$*^o1@0SNHcb6$LI=>Dh1eIQWc6v4z^K- zvegG2!b*Oj}vUZ5Tf=EKi+J1U2c_zF5dKw;%NBu$b zRQB>|lf!K#_VXpRcK(4;h^DP%l(29EjMj^CnY#Wi-Y)~SxRDXmXuw#z8~Z1S-*$38 zLppLlkdDe$mT8pO%yGOt*xcQ+?lAwF^tLaJA+1#WCd2~b-p{X*OQyv5Bm|L)*GiEk z57rbbEj>VJS=@t_TYOg2rkt=z3_82&V4f`bC@a{D+Zg`grU~|N%k*o$20eS5{8^e} z)JZj>eQk=ktke!xSWUlvtM%O`q*q9xQQ{YKaQ{i>@pK=x@fhiBl2zbi;g}~V9!8^W zm~li7F5GIz!vc7bvT_#|b&mUn!z*q+zXet0bsyzY*Csb|ZK4?C(q7!&kp zoDxbTcQ0OI9jM=SxZ%YUacfv5JZ4ZJ+T$0&C7V&(%GURT)_aOi}yW+1qG{q!`lpe`fL}cczr2R#Jo`{ zV7Qr=t-t%>p#V&&ovRotBBI=-C_|xg)8@h}a9BipbD1x*L*N0o=blUvEN0@gXA&(z z+ttO^Kgq1gBx3l3H52r=>Uml)GqP!LkkxhrQ|cf)R5zSi(;t;|3wP2V>XhObICL@R zhIakkn3iv}Up*C98v-iSWD|Tj%g$9a3>$XXotb(z2AN{Ji8THRgp*8=s^IR_&-5{8 znJJOFe~=5U8@CQMU^`A&o88Qo2fXBO^+&b@i+pBdPxaRUaXj-$H2V&h7F6Y zKT%XdFTDo3a;I~0iu7E{6Z!a?Ogf3@+Xy<5s4+%arv$&+Y8d?CK!uFLeaRSqQuN5o z!)zhN%VsM>F-e2;J@}MhdW&S!xeC&op*ljv%H~B*q#-XmNOg>g!s-H-^kT%TfO=QF z<$ejE;y-{rK>l^qusTCJ>XuMVfF%sttrsQ`h^p=j0fg1EPg|{JV_D^%DTQR(5H<=7 zD4|e~zE24aCWxM$FA5`NXiCRq$(|$t0j-ywME9}vgjAvW12I2f0t+YQW84|8yZ5XM zRnr{S@M}@8LHnFFBCT4u&yc)QAA0Yi!~Xa5Fh*64E6vM&e~&bjF0Sre20BTiMHeQZ zL+7>-4vGd!-sM}H#-Fn{BVG7gq+dennNDqDLsB}M*eodWO-VQ=+_7NE1{O#ns-b-< zsl7j{A1C3|b8r91_;C(h5yc1PLLs8Iic^dgSI)YZ9~M)~7UkAMso$F(IL^D}%-uc= z@R2Ch`GWzbNGa<|*#?)py+)C7^Ncs4+@egnQLMPpa!Yaj`*!kC!qdL?W&)aMQ_+pF za43TY8sbIsvpQqGbookT4*EZB}_a$`SVSI|l zyVtmMrhAmo?58~0r<(nv#RylH$Xs@*iC4|T*iZ9;Stft^P_Njv-hanxIbwAb5Bu3P z-%*{dLQRc+Excd!ue7zA>`R@7~!WmQ!i*^R{(EJlAZMRe<($!y3Rurs3eo zv%6DdK>qQoxOMvg_W}G$FaE)#py}n6Sqwb!smgmt`m+oLePy4F(T&*|FDN9XyYTcN z!QRL7^I5V8jAWoQmc$?#yrl7!i&5WzoN>9&KD-Hg zi2C%$VKy-r?as{4sPLelLbfl(R`%P_WD7j48uzri>eE^8O2`5}@YHFXnt6>~!vuY|{E$a&NKEhUhafHOzX8CGc zbR{^T+%l}#c9P`Ed(pa#Z7S4LL9e^?!;0{}70Y_gTSJ%f{Yy2T(otS*X3z$Q{My{? zW1tIPs#Md@cfw(?U}bo+d{`v8cYJh@cc;DqPOhTzo@E`oc>By&bth{MX~&lg5O@QS z4x58v=cUTlMg=AFwM69%6fGt2Q1J!UP(Mbl-(_!RV%j5m*(pTYTHZ6lxsv)ia=V>M-OR3 zDC5PyXtwr+<(n@|gc1EFkZDpw%y{TSNa7spS{#wuQN#OSvz48y6oPGiDIR<={e36s z3cfu8Q34oYlZkD8NZqqB53@5z#8^~MtMSd^dg)+~M-4Cu3b*{Ly*)^r-XG1-2*Grr ziz*TKzcIujlJ*>J;@;mb(S4d9G9?L39oZwMFZdn5DMDdoQ^hdF?jAar8_Y9k9fETJ zAh-;kx7s-0c!6^ly8agw@p)MP(%R{8-r6hrtt%T#1MdmnlhVD+$83bwiZe_y3%!(~ zon5b0(WRH#+G2X^eljf#10C|#0H~|qbfHREd*wxwZ4IR!U23$lBkSvZNHPn<@pHFt zV5oWKuy#~CUM1(cbwFDL*=xR8N<~dP48&7b6M+2nsGVOj>jBs>7%ovDTbVq6^=(d- z^af#~<>nznOlDJ%?o@PjP%6sv>tM)mHz9uNNO>>3HG*zNUXFUWsXyE?t4$E1&%XxW zg!l+V_!MdtgN<^}C-(q$fNvDHkN<-=p&m?1c2~6)=<32INeNn+p~a}jMW=y-SbOYs zh8LkCZZE{t0>@x>b3A;nzhf`dOjszvjo*KtGl}=UnPG&r&bM~#oV>{8+u$X}42TLd zzMrk;lT%}mtLo`9=PG@Dx{-M@@VQ~;Phqh5dw=`;mOrn|z4oHE#dD?6y{%{y&^P5# z>grQnj7I)<&$2IHMjXq!Yu%j9mF-DAOxPTmby2rlHJ1qe1cwLkZ}D1*7a00FI^M(R zg?^ndukWj2L^5ig@&2{dWliqBAF3WhCXPiz;+KiU=$v48{s7WIbPg)d?%r$y(3)5L zZH->fHrPMJr^*MDpmgof**k$_uJl@I`S0FFzZ=FM5Mj^NeK@vI8p_e3?oZ%My`T-{ z5f}54*Ta;z1k0zYj`?5bv1EC?+~s2bpUY5BnV}-P#_io@u#1&H(9yQRVDanj4HKuO za_2%x^+T8!lbA0SHxtpD>we-32MEex4Crc822qq@i?qhMqR>`IP@{^h%Q^X&sP+0*ePA{+G&26 zg2{|cw-($QoJ&lkV3ei(QsUH`*-vbx$(GBH9LzpF(47`r${>?_em!@f!ZsrUO!C0H*CEz z?bpUWH!MTS%3pwl@ox&W7|K)KkM)uw1#Ne-z|sq!y5fgw4edHTrg|z*Mlp9?r)q0a z4viW`F4lo=Pfz<%YosicDz2BlZjx1~Qz$eJY-?M&)9h8KWy4P#+1qpseC&nY_o{9m z=OhqQZ}!>8;X-cZbxy)3? z8Lx(6bzg@nf*aR8)izUJ4Q&X#>Ns7!4j{}3p|kR@cK3izm97doCi@nX$slWsSb{kI$nhvNFV@(%cgeUl9PsNQ z9(qvC3KSiePP}w7KFsSiJJk2M`53C4T%AelUqj>Hi3DpqNcvT7U#wuL5b{7TEMF%G zY(^ilfadm1*lM<&egU})$zbi23bhzubv|n~mrYDUKc7WrkW^ci?V#Yna(YIH_1pEgctc|et_14F zkFKq#2=@rH<}4Y(6UjrJs3b3sF`^s$Z0he?*`0Um)Xt+vtp$-0NBj_v;`(CyqY0WVJgQ?++e#(@VoSR=eK-uGWmVwC+>}Q~o5yz2w@e}AcvBv{xM|)Ypa34<{m9FB zMb3YDfUx-M#bLNtm9H*%)s|Z-NkGxVnkK!JLX<#J{v|QW@t-CW#@@c<+812V7SRO( zz*kUu^*tV>TC+oy0wn6+_%ge19w?;%go?^bd;Kv_0Bqb5aj4}%-khxgPr1PaMrKcU zyh|c^k!+f`v1vRJ{!;QXD`Z$t)I|?qNSM{?BY_H8caH!;VWMDas% ziS84J zKUVhvOFK5>uX!sCu;z3DA{nCWRm;Z*uA_Q#tGGagyvQrz>OYw5BFPxm%;g#T;R4#X zzcoSn6WS^pr9o{=r0u;$&qMOO>Kfg`NaF&13P6baCc+55QDJIfZ~`&#HFHP!T4zHX z$l`fLHmKoe0AuiqN)7H!!GZ9m>lBpr5nWTiXfK__l5^{dZ5lX(X}}NpO+~XOQ)CTJwHOF<{VL z#l~MZ%<4BsSF@SjiP^` z#86Xw2t?H+(1-s$j@FoYx=7(&{ z&ryeua1xSOSgkLu*1`kG%O~q&Yf4M(#y$`~D#5mC!jQ~b4E(Hhm1@Add85?c%FLPt_B`WAvUeAoDf%Tn0S&Wg z?^246tVBh7-R?Kt+L+A2IORZr&*|H6!udpSZOX;~2vt8#cyn#}#(UF^H{WJAWv!q! z{J<#~;_4jM1<)9TarJTQRjZq3t~!GwZ8WA$-??XS9n7#SpWYfr?}gK$IVe#;d*6D` zB4D}7o!frj`{EUYP6Y3X{>xG#ZytiAr0X@uzdk(YwkiCj zTB((!Jmwn2{L+zIUxPmU`ox&!L@DRB`C;B0HJ@11pAB3qzEE1_WLb*3riFQ{1XQ#g z_|F+KVN>0t>SZU!L2o}WK zM7ruGzs&^8ERY}U@k+Z=C0c4p&0Jk#m+toXM(?a!-}P+RDx(*=A&t8Mx1Ptbjyykb zPd>1m7FgQ(M7+fXcyw0lu!3z!!D$k_Jf`1A$JWo}f=z*IUTE~Q*JPIN-YZ22K-Y1dXTX<^ zdBjXe;tg%c9*@<0VlEqTsxqd_oTE5~@8!+1LZhZ7>FiOf?2%IVi@FyMKKk;JwQ6~s zRQs4|8V$62oJ(6-z*_hVQYSuxZs1!Bw63$4 zs&}#vkgrFuS=i`^@twj?uzG&5O-B>r?LQ!3dy#ll5dsiz;x7}pj7Yt!F<3GC)!QcxkgF!oz~#T$8g(#8i86%!ofORlxoDWk}|k+J}a7;^u`^s)Hh zXujkcNA)<|!i1NYG^Fvj-AA2y9wdm(Q;`rppgjzu#lnXJ!4&uafyyeps z6>EHb81aR%p0M%Sib_wjt3W&Vjh`w6B9MQ;l&}t>Flit}f*kMjUFChu%DLCNCAI6S z-JMR>m2T{54tI}a{$NLhE!{b}Wfo!Y@(<$db`aoocj#N+gwWy-fg?%QVihh;@?I(O z!IeQ@NG7>5Jhq;M0w%UC^=CwFbxaI($akH*v7&7ZJsNps=*m-- zm8N%9w)%hct6d1LCi>YzENrKli z8I5#WdDGh?!@kgXfopT}-m=~Mp+A~6xwC%5%5a#+qt6$3r>nkS;TT@|DOk7nC%No6 zDLUc=J8z+zChz5fdU}Q1*L|TOit94KX?qFl9873}TGr!C^X&t!Wnl%UZ>5Im}yp_n`0p?HVhm+MYq@_Qb3#!|%0do(q z{CASyY||#(X$S+7d{XP*4*v8;C?gB=Jza}ldD?cGo`kVgJw|dR1QW_^8LlqlcQ4Q6 z&g)5xg%&`#`Hem8bYF!y-9Y4rQV9wtq zCQ#Ab3JHfJ!pbX66v+wbw^wv+hgP2tq#{@~eztV%q$rjL_s<8vgmW)C!^hp$z$Lh8 z42R|CmVH3X({xrZ$0&aPci5nJK+GBtk5T2ZHj6RDuEZiJD<#P2)O^@ z?N`NvnVi2pxZ@5|U<$`lV7x2qGZlv(n}q=tSD=B&rPZe=Ju)_wlCoteQ&SDueX^;* z{z5^P1al>el4qQ1|#_1^wul?DzmBpcUM^NT8<&=c+7-yXkd5H)$calA*g#| z+l>&6ge9OgMZrGylQ4LqJ%-Wvc3oSL9 zymC(L9jA)rVq*i`0QSjQ|Mkq?@2e~j#{KV_=%t6v=+>cU6DH&Q~!_6k~dhpE%a;3 zHy!3OETky0P+*G@AipK8!2XGaG%gi7+yR8hgM=$JcoR=&#QNFOLsgwuJbQLccRYZ* zqeciY@G-BZ-0E*Tz)A4tmr#m{aX_2VV|lcldz+Vs23&@?;yf}elUepx&D=83=* z!$W*uhg=@)k2m82(xG>s$EglEW%@Fj;x3F;*`Ddo4EpW9MqiKqhrJEB$OLcLq5%0_+l!GfO$52u=>lW)CJS!Sx2-~axiLm!Fq#Hf@W8?F|MP5;;7 z9WX8&$Z}(net399#HTB&+BlhFeL0^L5A&x&_XLoGM*)sf$+P$1>l*%D_|BHJ_JJ|L z8DKE_F?4M?N}A^@SrLJ^JHr}&{?G6i)c_N#c^PFdt|X=${^y20ib87s#3gens$lLBq!|Y=VwiVt z?7s*0uf%dt`Ra^pogc^xL4WQk$9wreiv>85L15qsE0zy7L=XC>|Mw&LFIKpxn$cnV zAc~#k4nEzitYrUs6MLQsuqn+v0{@A@`(Ke&~9RMSM%Em zdFLblqZ9r;TlF}OlBrAPbzL$3x9n>RmShCX`1k$)|2*c{Ink_*qWvEZix;xK5n>V? z*a+u=0GPdyd>q&Wgtxtl|1nPf?X<{a zJYk;if_4M<;_GM5kiXFXeg^dSGsODB|J!GryuN&gTty~a~l^NXj}~tmNPP za@Njkfq!q0|FszZ4i^7wt9Gy;6=U>?Q;RLh|LP$K=kg(gemk2{RBz&c>1+>L-=ZmG zaE_O;k9!PB%-iAWAH1C>Pdxdp_~-cR*1>$j{~Cz8$P^m@=~{k73BYp8X@3vn|D(sc z^Rjb90=i3bbyJ?&9tzOG@Jgp2_m-|1hxUNyXzVWmuWRJ|-@o~P803GRY{*PppL+d?OmFi*Idf-b38=HXx;^Zg!K)Zmy z?k+lQ6Q~z{GHqE{{yPf+Ho*Q)^2>-GtIPl02!L%EE(?sA)B
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs001.html b/doc/pub/week42/html/._week42-bs001.html index be51bb83a..59127ba6d 100644 --- a/doc/pub/week42/html/._week42-bs001.html +++ b/doc/pub/week42/html/._week42-bs001.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs002.html b/doc/pub/week42/html/._week42-bs002.html index ecd3c51c9..c6623ae0b 100644 --- a/doc/pub/week42/html/._week42-bs002.html +++ b/doc/pub/week42/html/._week42-bs002.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs003.html b/doc/pub/week42/html/._week42-bs003.html index 55c376349..dcd01d0d6 100644 --- a/doc/pub/week42/html/._week42-bs003.html +++ b/doc/pub/week42/html/._week42-bs003.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs004.html b/doc/pub/week42/html/._week42-bs004.html index c0e0169e6..37db78730 100644 --- a/doc/pub/week42/html/._week42-bs004.html +++ b/doc/pub/week42/html/._week42-bs004.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs005.html b/doc/pub/week42/html/._week42-bs005.html index dedd4b8b1..62f983273 100644 --- a/doc/pub/week42/html/._week42-bs005.html +++ b/doc/pub/week42/html/._week42-bs005.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs006.html b/doc/pub/week42/html/._week42-bs006.html index 134e61953..f10997788 100644 --- a/doc/pub/week42/html/._week42-bs006.html +++ b/doc/pub/week42/html/._week42-bs006.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs007.html b/doc/pub/week42/html/._week42-bs007.html index aadf5059d..f36029253 100644 --- a/doc/pub/week42/html/._week42-bs007.html +++ b/doc/pub/week42/html/._week42-bs007.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs008.html b/doc/pub/week42/html/._week42-bs008.html index 200298551..9966971d8 100644 --- a/doc/pub/week42/html/._week42-bs008.html +++ b/doc/pub/week42/html/._week42-bs008.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs009.html b/doc/pub/week42/html/._week42-bs009.html index 7dc90011b..accfdbc6b 100644 --- a/doc/pub/week42/html/._week42-bs009.html +++ b/doc/pub/week42/html/._week42-bs009.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs010.html b/doc/pub/week42/html/._week42-bs010.html index e1512f11d..1cd03effb 100644 --- a/doc/pub/week42/html/._week42-bs010.html +++ b/doc/pub/week42/html/._week42-bs010.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs011.html b/doc/pub/week42/html/._week42-bs011.html index 72f64635e..bdb721fe0 100644 --- a/doc/pub/week42/html/._week42-bs011.html +++ b/doc/pub/week42/html/._week42-bs011.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs012.html b/doc/pub/week42/html/._week42-bs012.html index c419ade16..1876d220c 100644 --- a/doc/pub/week42/html/._week42-bs012.html +++ b/doc/pub/week42/html/._week42-bs012.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs013.html b/doc/pub/week42/html/._week42-bs013.html index 1b76b1762..b446fcd88 100644 --- a/doc/pub/week42/html/._week42-bs013.html +++ b/doc/pub/week42/html/._week42-bs013.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs014.html b/doc/pub/week42/html/._week42-bs014.html index ef02a49d8..8abe55087 100644 --- a/doc/pub/week42/html/._week42-bs014.html +++ b/doc/pub/week42/html/._week42-bs014.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs015.html b/doc/pub/week42/html/._week42-bs015.html index 07663628e..1746be476 100644 --- a/doc/pub/week42/html/._week42-bs015.html +++ b/doc/pub/week42/html/._week42-bs015.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs016.html b/doc/pub/week42/html/._week42-bs016.html index 44d62dfd1..4a9ec98ed 100644 --- a/doc/pub/week42/html/._week42-bs016.html +++ b/doc/pub/week42/html/._week42-bs016.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs017.html b/doc/pub/week42/html/._week42-bs017.html index b2f87e56e..31544c02e 100644 --- a/doc/pub/week42/html/._week42-bs017.html +++ b/doc/pub/week42/html/._week42-bs017.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -464,7 +465,7 @@ MathJax.Hub.Config({

     

     

     

    -

    Layout of a simple neural network with two input nodes, one hidden layer and one output node

    +

    Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node



    diff --git a/doc/pub/week42/html/._week42-bs018.html b/doc/pub/week42/html/._week42-bs018.html index 03567cd6e..84b0b581a 100644 --- a/doc/pub/week42/html/._week42-bs018.html +++ b/doc/pub/week42/html/._week42-bs018.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -466,16 +467,16 @@ MathJax.Hub.Config({

    The ouput layer

    -

    Finally, we have the ouput layer given by layer label \( (2) \) with output \( a^{(2)} \) and weights and biases to be determined given by the variables

    +

    We have the ouput layer given by layer label \( (2) \) with output \( a^{(2)} \) and weights and biases to be determined given by the variables

    $$ -w_{i}^{(2)}=\left\{w_{0}^{(2)},w_{1}^{(2)}\right\} \wedge b^{(2)}. +w_{i}^{(2)}=\left\{w_{1}^{(2)},w_{2}^{(2)}\right\} \wedge b^{(2)}. $$

    Our output is \( \tilde{y}=a^{(2)} \) and we define a generic cost function \( C(a^{(2)},y;\boldsymbol{\Theta}) \) where \( y \) is the target value (a scalar here). The parameters we need to optimize are given by

    $$ -\boldsymbol{\Theta}=\left\{w_{00}^{(1)},w_{01}^{(1)},w_{10}^{(1)},w_{11}^{(1)},w_{0}^{(2)},w_{1}^{(2)},b_0^{(1)},b_1^{(1)},b^{(2)}\right\}. +\boldsymbol{\Theta}=\left\{w_{11}^{(1)},w_{12}^{(1)},w_{21}^{(1)},w_{22}^{(1)},w_{1}^{(2)},w_{2}^{(2)},b_1^{(1)},b_2^{(1)},b^{(2)}\right\}. $$ diff --git a/doc/pub/week42/html/._week42-bs019.html b/doc/pub/week42/html/._week42-bs019.html index c2bc1a2d8..c41632cb4 100644 --- a/doc/pub/week42/html/._week42-bs019.html +++ b/doc/pub/week42/html/._week42-bs019.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -470,12 +471,12 @@ MathJax.Hub.Config({ The inputs to the first hidden layer are

    $$ -\begin{bmatrix}z_0^{(1)} \\ z_1^{(1)} \end{bmatrix}=\left(\begin{bmatrix}w_{00}^{(1)} & w_{01}^{(1)}\\ w_{10}^{(1)} &w_{11}^{(1)} \end{bmatrix}\right)^{T}\begin{bmatrix}a_0^{(0)} \\ a_1^{(0)} \end{bmatrix}+\begin{bmatrix}b_0^{(1)} \\ b_1^{(1)} \end{bmatrix}, +\begin{bmatrix}z_1^{(1)} \\ z_2^{(1)} \end{bmatrix}=\left(\begin{bmatrix}w_{11}^{(1)} & w_{12}^{(1)}\\ w_{21}^{(1)} &w_{22}^{(1)} \end{bmatrix}\right)^{T}\begin{bmatrix}a_1^{(0)} \\ a_2^{(0)} \end{bmatrix}+\begin{bmatrix}b_1^{(1)} \\ b_2^{(1)} \end{bmatrix}, $$

    with outputs

    $$ -\begin{bmatrix}a_0^{(1)} \\ a_1^{(1)} \end{bmatrix}=\begin{bmatrix}\sigma^{(1)}(z_0^{(1)}) \\ \sigma^{(1)}(z_1^{(1)}) \end{bmatrix}. +\begin{bmatrix}a_1^{(1)} \\ a_2^{(1)} \end{bmatrix}=\begin{bmatrix}\sigma^{(1)}(z_1^{(1)}) \\ \sigma^{(1)}(z_2^{(1)}) \end{bmatrix}. $$ diff --git a/doc/pub/week42/html/._week42-bs020.html b/doc/pub/week42/html/._week42-bs020.html index c9846baec..f193ffed7 100644 --- a/doc/pub/week42/html/._week42-bs020.html +++ b/doc/pub/week42/html/._week42-bs020.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -468,7 +469,7 @@ MathJax.Hub.Config({

    For the final output layer we have the inputs to the final activation function

    $$ -z^{(2)} = w_{0}^{(2)}a_0^{(1)} +w_{1}^{(2)}a_1^{(1)}+b^{(2)}, +z^{(2)} = w_{1}^{(2)}a_1^{(1)} +w_{2}^{(2)}a_2^{(1)}+b^{(2)}, $$

    resulting in the output

    diff --git a/doc/pub/week42/html/._week42-bs021.html b/doc/pub/week42/html/._week42-bs021.html index d3ab874da..e401afcfc 100644 --- a/doc/pub/week42/html/._week42-bs021.html +++ b/doc/pub/week42/html/._week42-bs021.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs022.html b/doc/pub/week42/html/._week42-bs022.html index 5d2782ec4..7af775cf9 100644 --- a/doc/pub/week42/html/._week42-bs022.html +++ b/doc/pub/week42/html/._week42-bs022.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -468,18 +469,18 @@ MathJax.Hub.Config({

    Using the chain rule we have the following expressions for say one of the weight parameters (it is easy to generalize to the other weight parameters)

    $$ -\frac{\partial C}{\partial w_{00}^{(1)}}=\frac{\partial C}{\partial a^{(2)}}\frac{\partial a^{(2)}}{\partial z^{(2)}} -\frac{\partial z^{(2)}}{\partial z_0^{(1)}}\frac{\partial z_0^{(1)}}{\partial w_{00}^{(1)}}= \delta^{(2)}\frac{\partial z^{(2)}}{\partial z_0^{(1)}}\frac{\partial z_0^{(1)}}{\partial w_{00}^{(1)}}, +\frac{\partial C}{\partial w_{11}^{(1)}}=\frac{\partial C}{\partial a^{(2)}}\frac{\partial a^{(2)}}{\partial z^{(2)}} +\frac{\partial z^{(2)}}{\partial z_1^{(1)}}\frac{\partial z_1^{(1)}}{\partial w_{11}^{(1)}}= \delta^{(2)}\frac{\partial z^{(2)}}{\partial z_1^{(1)}}\frac{\partial z_1^{(1)}}{\partial w_{11}^{(1)}}, $$

    which, noting that

    $$ -z^{(2)} =w_0^{(2)}a_0^{(1)}+w_1^{(2)}a_1^{(1)}+b^{(2)}, +z^{(2)} =w_1^{(2)}a_1^{(1)}+w_2^{(2)}a_2^{(1)}+b^{(2)}, $$

    allows us to rewrite

    $$ -\frac{\partial z^{(2)}}{\partial z_0^{(1)}}\frac{\partial z_0^{(1)}}{\partial w_{00}^{(1)}}=w_0^{(2)}\frac{\partial a_0^{(1)}}{\partial z_0^{(1)}}a_0^{(1)}. +\frac{\partial z^{(2)}}{\partial z_1^{(1)}}\frac{\partial z_1^{(1)}}{\partial w_{11}^{(1)}}=w_1^{(2)}\frac{\partial a_1^{(1)}}{\partial z_1^{(1)}}a_1^{(1)}. $$ diff --git a/doc/pub/week42/html/._week42-bs023.html b/doc/pub/week42/html/._week42-bs023.html index b54fdcd75..595d824e0 100644 --- a/doc/pub/week42/html/._week42-bs023.html +++ b/doc/pub/week42/html/._week42-bs023.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -467,17 +468,17 @@ MathJax.Hub.Config({

    Final expression

    Defining

    $$ -\delta_0^{(1)}=w_0^{(2)}\frac{\partial a_0^{(1)}}{\partial z_0^{(1)}}\delta^{(2)}, +\delta_1^{(1)}=w_1^{(2)}\frac{\partial a_1^{(1)}}{\partial z_1^{(1)}}\delta^{(2)}, $$

    we have

    $$ -\frac{\partial C}{\partial w_{00}^{(1)}}=\delta_0^{(1)}a_0^{(1)}. +\frac{\partial C}{\partial w_{11}^{(1)}}=\delta_1^{(1)}a_1^{(1)}. $$

    Similarly, we obtain

    $$ -\frac{\partial C}{\partial w_{01}^{(1)}}=\delta_0^{(1)}a_1^{(1)}. +\frac{\partial C}{\partial w_{12}^{(1)}}=\delta_1^{(1)}a_2^{(1)}. $$ diff --git a/doc/pub/week42/html/._week42-bs024.html b/doc/pub/week42/html/._week42-bs024.html index af2c49271..ea63e4fc7 100644 --- a/doc/pub/week42/html/._week42-bs024.html +++ b/doc/pub/week42/html/._week42-bs024.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -468,17 +469,17 @@ MathJax.Hub.Config({

    Similarly, we find

    $$ -\frac{\partial C}{\partial w_{10}^{(1)}}=\delta_1^{(1)}a_0^{(1)}, +\frac{\partial C}{\partial w_{21}^{(1)}}=\delta_2^{(1)}a_1^{(1)}, $$

    and

    $$ -\frac{\partial C}{\partial w_{11}^{(1)}}=\delta_1^{(1)}a_1^{(1)}, +\frac{\partial C}{\partial w_{22}^{(1)}}=\delta_2^{(1)}a_2^{(1)}, $$

    where we have defined

    $$ -\delta_1^{(1)}=w_1^{(2)}\frac{\partial a_1^{(1)}}{\partial z_1^{(1)}}\delta^{(2)}. +\delta_2^{(1)}=w_2^{(2)}\frac{\partial a_2^{(1)}}{\partial z_2^{(1)}}\delta^{(2)}. $$ diff --git a/doc/pub/week42/html/._week42-bs025.html b/doc/pub/week42/html/._week42-bs025.html index f0a2e0401..1a39902ee 100644 --- a/doc/pub/week42/html/._week42-bs025.html +++ b/doc/pub/week42/html/._week42-bs025.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -468,12 +469,12 @@ MathJax.Hub.Config({

    For the sake of completeness, we list the derivatives of the biases, which are

    $$ -\frac{\partial C}{\partial b_{0}^{(1)}}=\delta_0^{(1)}, +\frac{\partial C}{\partial b_{1}^{(1)}}=\delta_1^{(1)}, $$

    and

    $$ -\frac{\partial C}{\partial b_{1}^{(1)}}=\delta_1^{(1)}. +\frac{\partial C}{\partial b_{2}^{(1)}}=\delta_2^{(1)}. $$

    As we will see below, these expressions can be generalized in a more compact form.

    diff --git a/doc/pub/week42/html/._week42-bs026.html b/doc/pub/week42/html/._week42-bs026.html index 79e07a6cc..f00c2656b 100644 --- a/doc/pub/week42/html/._week42-bs026.html +++ b/doc/pub/week42/html/._week42-bs026.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs027.html b/doc/pub/week42/html/._week42-bs027.html index 87cc9972a..89668d155 100644 --- a/doc/pub/week42/html/._week42-bs027.html +++ b/doc/pub/week42/html/._week42-bs027.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs028.html b/doc/pub/week42/html/._week42-bs028.html index 9feba3653..c2b11af09 100644 --- a/doc/pub/week42/html/._week42-bs028.html +++ b/doc/pub/week42/html/._week42-bs028.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs029.html b/doc/pub/week42/html/._week42-bs029.html index 9886f5380..45bfd2bbd 100644 --- a/doc/pub/week42/html/._week42-bs029.html +++ b/doc/pub/week42/html/._week42-bs029.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs030.html b/doc/pub/week42/html/._week42-bs030.html index 26a71dc2b..04978b68b 100644 --- a/doc/pub/week42/html/._week42-bs030.html +++ b/doc/pub/week42/html/._week42-bs030.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs031.html b/doc/pub/week42/html/._week42-bs031.html index 38241757c..40e18abe1 100644 --- a/doc/pub/week42/html/._week42-bs031.html +++ b/doc/pub/week42/html/._week42-bs031.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs032.html b/doc/pub/week42/html/._week42-bs032.html index 8d1802082..4090147f3 100644 --- a/doc/pub/week42/html/._week42-bs032.html +++ b/doc/pub/week42/html/._week42-bs032.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs033.html b/doc/pub/week42/html/._week42-bs033.html index ff1904224..ae79e8047 100644 --- a/doc/pub/week42/html/._week42-bs033.html +++ b/doc/pub/week42/html/._week42-bs033.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs034.html b/doc/pub/week42/html/._week42-bs034.html index 4b22fd5c8..76f262307 100644 --- a/doc/pub/week42/html/._week42-bs034.html +++ b/doc/pub/week42/html/._week42-bs034.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs035.html b/doc/pub/week42/html/._week42-bs035.html index a6693827a..b2d367ad6 100644 --- a/doc/pub/week42/html/._week42-bs035.html +++ b/doc/pub/week42/html/._week42-bs035.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs036.html b/doc/pub/week42/html/._week42-bs036.html index 4c0068c55..6d921843a 100644 --- a/doc/pub/week42/html/._week42-bs036.html +++ b/doc/pub/week42/html/._week42-bs036.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs037.html b/doc/pub/week42/html/._week42-bs037.html index 187c57639..d1fe17793 100644 --- a/doc/pub/week42/html/._week42-bs037.html +++ b/doc/pub/week42/html/._week42-bs037.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs038.html b/doc/pub/week42/html/._week42-bs038.html index e31f59398..2e20d7b0d 100644 --- a/doc/pub/week42/html/._week42-bs038.html +++ b/doc/pub/week42/html/._week42-bs038.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs039.html b/doc/pub/week42/html/._week42-bs039.html index 4e00e8c88..e4d3d3833 100644 --- a/doc/pub/week42/html/._week42-bs039.html +++ b/doc/pub/week42/html/._week42-bs039.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs040.html b/doc/pub/week42/html/._week42-bs040.html index 1d13e5439..298d6005a 100644 --- a/doc/pub/week42/html/._week42-bs040.html +++ b/doc/pub/week42/html/._week42-bs040.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • diff --git a/doc/pub/week42/html/._week42-bs041.html b/doc/pub/week42/html/._week42-bs041.html index 19c1dd6ad..4c0d90908 100644 --- a/doc/pub/week42/html/._week42-bs041.html +++ b/doc/pub/week42/html/._week42-bs041.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({
  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -464,22 +465,23 @@ MathJax.Hub.Config({

     

     

     

    -

    Setting up the back propagation algorithm

    +

    Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations

    -

    The four equations provide us with a way of computing the gradient of the cost function. Let us write this out in the form of an algorithm.

    +
    +
    + +
      +
    1. Set up your inputs and outputs (scalars, vectors, matrices or higher-order arrays)
      2. +
    2. Define the number of hidden layers and hidden nodes
      3. +
    3. Define activation functions for hidden layers and output layers
      4. +
    4. Define optimizer (plan learning rate, momentum, ADAgrad, RMSprop, ADAM etc) and array of initial learning rates
      5. +
    5. Define cost function and possible regularization terms with hyperparameters
      6. +
    6. Initialize weights and biases
      7. +
    7. Fix number of iterations for the feed forward part and back propagation part
      8. +
    +
    +
    -

    First, we set up the input data \( \boldsymbol{x} \) and the activations -\( \boldsymbol{z}_1 \) of the input layer and compute the activation function and -the pertinent outputs \( \boldsymbol{a}^1 \). -

    - -

    Secondly, we perform then the feed forward till we reach the output -layer and compute all \( \boldsymbol{z}_l \) of the input layer and compute the -activation function and the pertinent outputs \( \boldsymbol{a}^l \) for -\( l=1,2,3,\dots,L \). -

    - -

    Notation: The first hidden layer has \( l=1 \) as label and the final output layer has \( l=L \).

    diff --git a/doc/pub/week42/html/._week42-bs042.html b/doc/pub/week42/html/._week42-bs042.html index bf46208ac..51b2a3657 100644 --- a/doc/pub/week42/html/._week42-bs042.html +++ b/doc/pub/week42/html/._week42-bs042.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({

  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -464,18 +465,22 @@ MathJax.Hub.Config({

     

     

     

    -

    Setting up the back propagation algorithm, part 2

    +

    Setting up the back propagation algorithm, part 1

    -

    Thereafter we compute the ouput error \( \boldsymbol{\delta}^L \) by computing all

    -$$ -\delta_j^L = \sigma'(z_j^L)\frac{\partial {\cal C}}{\partial (a_j^L)}. -$$ +

    The four equations provide us with a way of computing the gradients of the cost function. Let us write this out in the form of an algorithm.

    -

    Then we compute the back propagate error for each \( l=L-1,L-2,\dots,1 \) as

    -$$ -\delta_j^l = \sum_k \delta_k^{l+1}w_{kj}^{l+1}\sigma'(z_j^l). -$$ +

    First, we set up the input data \( \boldsymbol{x} \) and the activations +\( \boldsymbol{z}_1 \) of the input layer and compute the activation function and +the pertinent outputs \( \boldsymbol{a}^1 \). +

    +

    Secondly, we perform then the feed forward till we reach the output +layer and compute all \( \boldsymbol{z}_l \) of the input layer and compute the +activation function and the pertinent outputs \( \boldsymbol{a}^l \) for +\( l=1,2,3,\dots,L \). +

    + +

    Notation: The first hidden layer has \( l=1 \) as label and the final output layer has \( l=L \).

    diff --git a/doc/pub/week42/html/._week42-bs043.html b/doc/pub/week42/html/._week42-bs043.html index 164217d4e..416827e4a 100644 --- a/doc/pub/week42/html/._week42-bs043.html +++ b/doc/pub/week42/html/._week42-bs043.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({

  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -464,23 +465,18 @@ MathJax.Hub.Config({

     

     

     

    -

    Setting up the Back propagation algorithm, part 3

    - -

    Finally, we update the weights and the biases using gradient descent -for each \( l=L-1,L-2,\dots,1 \) (the first hidden layer) and update the weights and biases -according to the rules -

    +

    Setting up the back propagation algorithm, part 2

    +

    Thereafter we compute the ouput error \( \boldsymbol{\delta}^L \) by computing all

    $$ -w_{ij}^l\leftarrow = w_{ij}^l- \eta \delta_j^la_i^{l-1}, +\delta_j^L = \sigma'(z_j^L)\frac{\partial {\cal C}}{\partial (a_j^L)}. $$ - +

    Then we compute the back propagate error for each \( l=L-1,L-2,\dots,1 \) as

    $$ -b_j^l \leftarrow b_j^l-\eta \frac{\partial {\cal C}}{\partial b_j^l}=b_j^l-\eta \delta_j^l, +\delta_j^l = \sum_k \delta_k^{l+1}w_{kj}^{l+1}\sigma'(z_j^l). $$ -

    with \( \eta \) being the learning rate.

    diff --git a/doc/pub/week42/html/._week42-bs044.html b/doc/pub/week42/html/._week42-bs044.html index 27689cfbc..612fa717b 100644 --- a/doc/pub/week42/html/._week42-bs044.html +++ b/doc/pub/week42/html/._week42-bs044.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({

  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -464,14 +465,13 @@ MathJax.Hub.Config({

     

     

     

    -

    Updating the gradients

    +

    Setting up the Back propagation algorithm, part 3

    -

    With the back propagate error for each \( l=L-1,L-2,\dots,1 \) as

    -$$ -\delta_j^l = \sum_k \delta_k^{l+1}w_{kj}^{l+1}\sigma'(z_j^l), -$$ +

    Finally, we update the weights and the biases using gradient descent +for each \( l=L-1,L-2,\dots,1 \) (the first hidden layer) and update the weights and biases +according to the rules +

    -

    we update the weights and the biases using gradient descent for each \( l=L-1,L-2,\dots,1 \) and update the weights and biases according to the rules

    $$ w_{ij}^l\leftarrow = w_{ij}^l- \eta \delta_j^la_i^{l-1}, $$ @@ -481,6 +481,7 @@ $$ b_j^l \leftarrow b_j^l-\eta \frac{\partial {\cal C}}{\partial b_j^l}=b_j^l-\eta \delta_j^l, $$ +

    with \( \eta \) being the learning rate.

    diff --git a/doc/pub/week42/html/._week42-bs045.html b/doc/pub/week42/html/._week42-bs045.html index c7013865a..2a9abcd83 100644 --- a/doc/pub/week42/html/._week42-bs045.html +++ b/doc/pub/week42/html/._week42-bs045.html @@ -81,10 +81,10 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'simple-neural-network-and-the-back-propagation-equations'), ('Layout of a simple neural network with two input nodes, one ' - 'hidden layer and one output node', + 'hidden layer with two hidden noeds and one output node', 2, None, - 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-and-one-output-node'), + 'layout-of-a-simple-neural-network-with-two-input-nodes-one-hidden-layer-with-two-hidden-noeds-and-one-output-node'), ('The ouput layer', 2, None, 'the-ouput-layer'), ('Compact expressions', 2, None, 'compact-expressions'), ('Output layer', 2, None, 'output-layer'), @@ -149,10 +149,15 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'using-the-chain-rule-and-summing-over-all-k-entries'), - ('Setting up the back propagation algorithm', + ('Setting up the back propagation algorithm and algorithm for a ' + 'feed forward NN, initalizations', 2, None, - 'setting-up-the-back-propagation-algorithm'), + 'setting-up-the-back-propagation-algorithm-and-algorithm-for-a-feed-forward-nn-initalizations'), + ('Setting up the back propagation algorithm, part 1', + 2, + None, + 'setting-up-the-back-propagation-algorithm-part-1'), ('Setting up the back propagation algorithm, part 2', 2, None, @@ -162,17 +167,12 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d None, 'setting-up-the-back-propagation-algorithm-part-3'), ('Updating the gradients', 2, None, 'updating-the-gradients'), - ('Activation functions', 3, None, 'activation-functions'), + ('Activation functions', 2, None, 'activation-functions'), ('Activation functions, Logistic and Hyperbolic ones', 3, None, 'activation-functions-logistic-and-hyperbolic-ones'), - ('Relevance', 3, None, 'relevance'), - ('Fine-tuning neural network hyperparameters', - 2, - None, - 'fine-tuning-neural-network-hyperparameters'), - ('Hidden layers', 2, None, 'hidden-layers'), + ('Relevance', 2, None, 'relevance'), ('Vanishing gradients', 2, None, 'vanishing-gradients'), ('Exploding gradients', 2, None, 'exploding-gradients'), ('Is the Logistic activation function (Sigmoid) our choice?', @@ -201,6 +201,11 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d 2, None, 'more-on-activation-functions-output-layers'), + ('Fine-tuning neural network hyperparameters', + 2, + None, + 'fine-tuning-neural-network-hyperparameters'), + ('Hidden layers', 2, None, 'hidden-layers'), ('Batch Normalization', 2, None, 'batch-normalization'), ('Dropout', 2, None, 'dropout'), ('Gradient Clipping', 2, None, 'gradient-clipping'), @@ -219,10 +224,6 @@ doconce format html week42.do.txt --html_style=bootstrap --pygments_html_style=d ('Limitations of NNs', 2, None, 'limitations-of-nns'), ('Homogeneous data', 2, None, 'homogeneous-data'), ('More limitations', 2, None, 'more-limitations'), - ('Setting up the back-propagation algorithm', - 2, - None, - 'setting-up-the-back-propagation-algorithm'), ('Setting up a Multi-layer perceptron model for classification', 2, None, @@ -364,7 +365,7 @@ MathJax.Hub.Config({

  • The training
  • Code example
  • Simple neural network and the back propagation equations
    • -
  • Layout of a simple neural network with two input nodes, one hidden layer and one output node
    • +
  • Layout of a simple neural network with two input nodes, one hidden layer with two hidden noeds and one output node
  • The ouput layer
  • Compact expressions
  • Output layer
    • @@ -388,35 +389,35 @@ MathJax.Hub.Config({
  • Bringing it together
  • Final back propagating equation
  • Using the chain rule and summing over all \( k \) entries
    • -
  • Setting up the back propagation algorithm
    • -
  • Setting up the back propagation algorithm, part 2
    • -
  • Setting up the Back propagation algorithm, part 3
    • -
  • Updating the gradients
    • -
  •    Activation functions
    • -
  •    Activation functions, Logistic and Hyperbolic ones
    • -
  •    Relevance
    • -
  • Fine-tuning neural network hyperparameters
    • -
  • Hidden layers
    • -
  • Vanishing gradients
    • -
  • Exploding gradients
    • -
  • Is the Logistic activation function (Sigmoid) our choice?
    • -
  • Logistic function as the root of problems
    • -
  • The derivative of the Logistic funtion
    • -
  • Insights from the paper by Glorot and Bengio
    • -
  • The RELU function family
    • -
  • ELU function
    • -
  • Which activation function should we use?
    • -
  • More on activation functions, output layers
    • -
  • Batch Normalization
    • -
  • Dropout
    • -
  • Gradient Clipping
    • -
  • A top-down perspective on Neural networks
    • -
  • More top-down perspectives
    • -
  • Limitations of supervised learning with deep networks
    • -
  • Limitations of NNs
    • -
  • Homogeneous data
    • -
  • More limitations
    • -
  • Setting up the back-propagation algorithm
    • +
  • Setting up the back propagation algorithm and algorithm for a feed forward NN, initalizations
    • +
  • Setting up the back propagation algorithm, part 1
    • +
  • Setting up the back propagation algorithm, part 2
    • +
  • Setting up the Back propagation algorithm, part 3
    • +
  • Updating the gradients
    • +
  • Activation functions
    • +
  •    Activation functions, Logistic and Hyperbolic ones
    • +
  • Relevance
    • +
  • Vanishing gradients
    • +
  • Exploding gradients
    • +
  • Is the Logistic activation function (Sigmoid) our choice?
    • +
  • Logistic function as the root of problems
    • +
  • The derivative of the Logistic funtion
    • +
  • Insights from the paper by Glorot and Bengio
    • +
  • The RELU function family
    • +
  • ELU function
    • +
  • Which activation function should we use?
    • +
  • More on activation functions, output layers
    • +
  • Fine-tuning neural network hyperparameters
    • +
  • Hidden layers
    • +
  • Batch Normalization
    • +
  • Dropout
    • +
  • Gradient Clipping
    • +
  • A top-down perspective on Neural networks
    • +
  • More top-down perspectives
    • +
  • Limitations of supervised learning with deep networks
    • +
  • Limitations of NNs
    • +
  • Homogeneous data
    • +
  • More limitations
  • Setting up a Multi-layer perceptron model for classification
  • Defining the cost function
  • Example: binary classification problem
    • @@ -464,20 +465,24 @@ MathJax.Hub.Config({

     

     

     

    -

    Activation functions

    +

    Updating the gradients

    + +

    With the back propagate error for each \( l=L-1,L-2,\dots,1 \) as

    +$$ +\delta_j^l = \sum_k \delta_k^{l+1}w_{kj}^{l+1}\sigma'(z_j^l), +$$ + +

    we update the weights and the biases using gradient descent for each \( l=L-1,L-2,\dots,1 \) and update the weights and biases according to the rules

    +$$ +w_{ij}^l\leftarrow = w_{ij}^l- \eta \delta_j^la_i^{l-1}, +$$ + + +$$ +b_j^l \leftarrow b_j^l-\eta \frac{\partial {\cal C}}{\partial b_j^l}=b_j^l-\eta \delta_j^l, +$$ -

    A property that characterizes a neural network, other than its -connectivity, is the choice of activation function(s). As described -in, the following restrictions are imposed on an activation function -for a FFNN to fulfill the universal approximation theorem -

    -
      -
    • Non-constant
      • -
    • Bounded
      • -
    • Monotonically-increasing
      • -
    • Continuous
      • -

  • The training