From 3bb72da70ff2d87f8c01e212b77b2d7fe80987f1 Mon Sep 17 00:00:00 2001 From: Lars Bogner Date: Sat, 4 Oct 2025 11:06:05 +0200 Subject: [PATCH] Finalize report --- report/chapters/conclusion.tex | 2 +- report/chapters/introduction.tex | 4 ++-- report/chapters/perspective.tex | 4 +++- report/main.pdf | Bin 807658 -> 808182 bytes report/main.tex | 2 +- 5 files changed, 7 insertions(+), 5 deletions(-) diff --git a/report/chapters/conclusion.tex b/report/chapters/conclusion.tex index 5767c2a..389e737 100644 --- a/report/chapters/conclusion.tex +++ b/report/chapters/conclusion.tex @@ -1 +1 @@ -It was shown that in the context of big data and complex models to describe relations in the data there are many challenges to overcome. The choice of model, on what metric to optimize the model -- i.e. the cost function -- and how to perform the optimization are all crucial choices that can have a significant impact on the performance of the model. The use of regularization techniques can help to prevent overfitting and improve the generalization performance of the model. The number of degrees of freedom in the model directly influcences the bias-variance trade-off and only with a good balance between the two can a model perform well on unseen data. Finally, the choice of optimization algorithm can have a significant impact on the computational cost of the models training process. Using good optimization algorithms can help to speed up the convergence of the models parameters. Combining this with other innovative techniques such as stochastic gradient descent can help to further reduce the computational cost of training complex models on large datasets. \ No newline at end of file +It was shown that in the context of big data and complex models to describe relations in the data there are many challenges to overcome. The choice of model, on what metric to optimize the model -- i.e. the cost function -- and how to perform the optimization are all crucial choices that can have a significant impact on the performance of the model. The use of regularization techniques can help to prevent overfitting and improve the generalization performance of the model. The number of degrees of freedom in the model directly influences the bias-variance trade-off and only with a good balance between the two can a model perform well on unseen data. Finally, the choice of optimization algorithm can have a significant impact on the computational cost of the models training process. Using good optimization algorithms can help to speed up the convergence of the models parameters. Combining this with other innovative techniques such as stochastic gradient descent can help to further reduce the computational cost of training complex models on large datasets. \ No newline at end of file diff --git a/report/chapters/introduction.tex b/report/chapters/introduction.tex index 55f5822..06a604d 100644 --- a/report/chapters/introduction.tex +++ b/report/chapters/introduction.tex @@ -1,7 +1,7 @@ Optimizing numerical models to model a set of data has been a challenge of the natural sciences for many decades. But with the introduction of artificial neural networks in all disciplines over the last years a special focus on those optimization techniques is relevant. -In this paper different methods of such optimization are studied in the context of large datasets and big data with possibly noise-affected input data. To render the best possible fits a multitude of cost functions, minimization algorithms and other approaches to reduce the computation associated with the optimization process are studied and evaluated. Reducing the computational cost is also important to improve the economical and ecological footprint of training large models on big data, as the use of datadriven methods in all areas of life is increasing rapidly. To measure the possible improvements with state-of-the-art methods, the duration of optimization and the quality of the fit are evaluated. Assuming constant power consumption of the hardware, a reduction in computation time directly translates to a reduction in energy consumption and thus a smaller ecological and economical footprint. +In this paper different methods of such optimization are studied in the context of large datasets and big data with possibly noise-affected input data. To render the best possible fits a multitude of cost functions, minimization algorithms and other approaches to reduce the computation associated with the optimization process are studied and evaluated. Reducing the computational cost is also important to improve the economical and ecological footprint of training large models on big data, as the use of data-driven methods in all areas of life is increasing rapidly. To measure the possible improvements with state-of-the-art methods, the duration of optimization and the quality of the fit are evaluated. Assuming constant power consumption of the hardware, a reduction in computation time directly translates to a reduction in energy consumption and thus a smaller ecological and economical footprint. -In a first step, the dependence of the bias-variance tradeoff on the number of degrees of freedom in a model is studied. For this purpose, polynomial models of different degrees are fitted to data generated from the Runge function with significant noise. The results are then compared to different Ridge and Lasso models to study the effect of regularization on the bias-variance tradeoff. In the context of big data, the performance of different numerical minimization algorithms is studied. For this purpose, the performance of gradient descent, stochastic gradient descent and different optimization algorithms such as Adam are compared in the context of Ridge and Lasso regression. Furthermore the possible advantages and disadvantages of stochastic gradient descent techniques are evaluated. Finally, the effect of resampling methods such as bootstrapping and k-fold cross-validation on the bias-variance tradeoff is studied. +In a first step, the dependence of the bias-variance tradeoff on the number of degrees of freedom in a model is studied. For this purpose, polynomial models of different degrees are fitted to data generated from the Runge function with significant noise. The results are then compared to different Ridge and Lasso models to study the effect of regularization on the bias-variance tradeoff. In the context of big data, the performance of different numerical minimization algorithms is studied. For this purpose, the performance of gradient descent, stochastic gradient descent and different optimization algorithms such as Adam are compared in the context of Ridge and Lasso regression. Furthermore, the possible advantages and disadvantages of stochastic gradient descent techniques are evaluated. Finally, the effect of resampling methods such as bootstrapping and k-fold cross-validation on the bias-variance tradeoff is studied. In the following section the theoretical background of the methods used in this paper is presented. In \cref{sec:methods} the implementation of these methods is discussed. The results of applying these methods to different datasets are presented in \cref{sec:results}. Finally, a conclusion is drawn in \cref{sec:conclusion} and an outlook on possible future work is given in \cref{sec:perspective}. diff --git a/report/chapters/perspective.tex b/report/chapters/perspective.tex index bc66453..d0e4692 100644 --- a/report/chapters/perspective.tex +++ b/report/chapters/perspective.tex @@ -1 +1,3 @@ -With the increasing availability of large datasets and the growing complexity of models, there is an increasing need for efficient optimization techniques that can handle the computational challenges associated with big data. Especially in the context of deep learning, where models can have millions of parameters and require large amounts of data to train, the choice of optimization algorithm can have a significant impact on the performance of the model. As to limit the computational footprint of such datadriven methods, the search for more efficient optimization techniques is an ongoing field of research. There will also be further advances in the methods to choose hyperparameters of model training to make the usage of numerical models easier and more efficient to use. \ No newline at end of file +With the increasing availability of large datasets and the growing complexity of models, there is an increasing need for efficient optimization techniques that can handle the computational challenges associated with big data. Especially in the context of deep learning, where models can have millions of parameters and require large amounts of data to train, the choice of optimization algorithm can have a significant impact on the performance of the model. As to limit the computational footprint of such data-driven methods, the search for more efficient optimization techniques is an ongoing field of research. There will also be further advances in the methods to choose hyperparameters of model training to make the usage of numerical models easier and more efficient to use. + +As the scope of this paper was limited to the study of a simple toy dataset with simple models, the results cannot be directly transferred to more complex models and datasets, and need to be researched accordingly. However, the general trends observed in this study can be expected to hold true for more complex models and datasets as well. The choice of optimization algorithm and cost function can have a significant impact on the performance of the model, and reducing the computational cost of training is an important consideration in the context of big data. \ No newline at end of file diff --git a/report/main.pdf b/report/main.pdf index fc4cc8a8e70f984ed0fd6d9cdcec401a34c5dac7..b05a243c7636aa8490b10b6aef702e0aa4bca7af 100644 GIT binary patch delta 19671 zcmV(~K+nJGr7`xYF_0z#H!_!jH3TS;P!)f&EY~Ksq>|Q3xsp6!Nsx;O1ZW&uZ+?CH z>+Zq9#k%rB40`Uq`vd)((~W=g(|7*W^Ud9NuilyXCa^nW#7pMZ<{Renn7OPZ?=Dd zR}_c8ZvK7uH`?9aWVjPE_k z+ad4s?I!RJZTW2zW?oV2sG@&pHd~?`I-$6QdbfD9;5WLOqxA3w!?|e7eO~E$v=|W& zMdg-h^Z-S-Z_9h$s=(E7nQUjlP~4?0D&B#nSwHg1TAnyJu`Uj+_YdNz5cz*-GHTxz zbocrs?>VmaZ4+f)URL>iS(Ux=0KGUip2Yf7F)iN}y>7}Xw%Fj*Hdx?s++p9GFIDk} zTX%MkNbbR#m~hUB?}-vIL)$_+OQ>Wg?Yvgn{4S(?HL0xhut-5F{^bYpzATvUz0j7@ z!SKG=gAEUL`TJ1lCjCRMIXF&nrW`#s=OVJ3r(VCI%iIl`IR! zV(A{#?mD^kzRY`^9#bA!Y-rt=CuyiqZJc`VHfiP!EqHTo+Twc*u(TGBJBem0Cxy1W zFDRtKq=0%py4`b`-17p381=p#=y;X9Ks)gy##;6-pgr(N>#4GKz0iLzyh_t`<$a}V zO+sESH!AzIt*&aO%SYaAEgSU@rFxu8evwUE9bD4M4Cze`k?KLzX0fqOBc*#TzHSgb zpdndyttYnliU9o>%kzvk^*o#`FDV>va6V1eNG)jzj2X$&>1TpQve55w*7_cO$ ztzzkP_O)1P1hfH{L3b{sGrPK~FNs$=3VgtG-?*iYL%nAo1~txN?_Aas&z}n@uTD)% z%s7K@S*q>_vVu7%?Q&_GT+EJ*nYy86=xd9^urHLx;>yyw(Mo?$&PGXmc1DR&DuemN zg@QciKOI9>DLQZ_j75J;7)|pf{o=MI(yefwcMeHn1!;G4mhgplKD~BS&B=Ad``b%3n z_lt2*0zCURCt-i%W_VN6K=Z88l>;KrsO8C)5vuw^VPH8rW;|bap_p)xBe<<3ls}gc zhjwT-yb6V?WoS3PW!uIx=R64k4-rRbtj+YS%44CU$5~Tg!h4%v$^+ResXoLSMc%Kj zlVU8{87UH?!%_w*CcbPRZ`W9IYv5;Z_B0x)g;}cPw3%p_PtnexU3{AMp(v+CQiv%lsQ9c{* zGD;D*LArkn1GgRG%5`Mv+{5@vHb!i%iA&^#eN4$i{3-gAS8h7 zybkqC^)OWqHlbWQoD^$3d8j+SpYm7CUltdE2{M1PoUR8pI{h=9q}QZJa-**2WP}d& zQ=1>AWDTporMzGaW_9sTMTeW7!H`QOHb$%bvM}tx#)jJ~ zedA6F(>(Ur{1HbI_f5xiF%upYPHMr=Tco>Zy@YZlGc-XqobP9sqWnnPwaux`&u2Y( z!Wc=)lvr@?#63CqbARWC&o}{gi*sMtx+Ck}Qk9;Odw? zr>f(W1TU?7bmAX5EVWd}oW+tuI!ma>_E#Oj2ROt-S?BGODoDXDp_jMMzYnCp%zsO3 z$A4nr8)<~x*}=U zr1mm-%|qgLuV;B|ngqU=ia2(Ozf{H_os*yk7Zn7K$o|*n7!d07%pAYyiHJ%2YWNeA z%m*BQGSg?FH`-+;90R<+x)~)BCJS`45jQ)cPNe*NZCRahwd^vX!98hFLpq)YQuLB0NFLY6XXQ%llPlREuu`Bamz?p ztniDa+*{xmy(IO_;g?hpw|K0p5>CH=KoQI@7Q`h5;=D zIpH!0?+n6#z}`O`D@_cZKp1pr9j`K+g|JfHBU5^k9)V@%7!0V2 z!Pp7XO9*c#w``DXf$T6CYyEkFH{V`u5=s<>{?!@i2#m_0$DuN4=khqSQyEch8ITy8 z4t)g>(3+z%Q7ov;cI_l{SkYggGk_L_q1Jck4AT^yfzSm9^`#o%s|z|~UcUMQs|f*~ z2v(DhSPh6GpU?p&XecGz#RBqe&DImOn{BCbHK)7f*5)Y z*ppBS7k@yU#DO?$U7!p5e68e1ucMt0%FsD(!|T=x4-0OyDqa~4|E(@Z!5p>hBvG$F z2XpY|a87ETC12Et)ak5RgG;Fs6wrwR13+ixByx>3@&@*BitT{F*p7UoNd?FuBJa9z zI+#*wuYft}1m?_^Q!po06pd%C@@9tc*mOd8EPo?Blx5wa-AM15h7Rnutn*OR5XU*e zs6@IFA<8KD2=+S*kzQlH?wQxv^ z-G5CM5t`8}gLjWE0*8iy`EEF0)CbT&1_+IsX~Ouorgoqm3QaF4D00;UA_uG>m`2^BIwBo{MT5;;AdUG$c%{KyFKMxa z$WUKl>za4t_kPT6c;r!@hvL*0EUy{yNTfq^RyZ{7r30kc&W+!R-JAcjz zTIrH}3$4q58#>p)I!d2SRN<>9$i)Hhiwa?a!&Mc90pumy?Yc^h5hj^k*rsZ=RxMdS zS|%EWG-u+0C#YpU4mh^V9qW7nH9id}A69mE2?(7xcAvjO`ZhAAI4Ay?s`s#a&0A#< zXI{)jU%vNajn+X%c;PFcK_}5Y$bWKI`Ea4JxIm(onatQHmD$888uO_L3)i8nZY=vd8B!5@PW@ zDY~Ufi%_iS7BPVrhC9^6R%&g5#&q)P$z(<0Vra${6Gzbc#;h%2z z1Thq$HPMTO@rtA4PldCHZ_EgLh$sMci#xc=S<)$GF|oooXwX3`PSWK+P9SxK%hBco z`^6_dBHT;iJAUE=O@IH0vYM%gjRZ^HvlUPobWo=@8kBQyb4X@#d*snbr0NjV`5C?( zyF_;Z47}y(w=V|+`odSb&pd1?dihK-_Kg3$)9KLV>3Om~5AwxEa(;5fck5!EVz>Np zho7WK)sy^vz>M34=6btO|6>%I6~7nipNj+o;#Jqnf5$rbDSxj0LrwYTqNf+fIsV-V z?aK1}R8o8Pd=5dE6+3wzTK-GrV?A!PW^SXv$Q*9J$WxpoH{{t2eWk|tRVmstf=1%~ zKspXR7pZ7FxKrL(gAD$m`Hca~munzc)yNPsy)SH4&;Hc<k4IVWOHHpk8i&GPMJ*^m$}L}kI$PdE8;ZOn_ZH{MV4(I_nUu2<-{>|Uv&kb#1yREk$VQG@h z4zSD(VF~ADNu2BoM)14la2f2wR_}IEnmlZ?yo|nXs&V__AuFP4XsWK3;k~W<)A?bm zc2PezQ`2`0@1OZQ9cALlRy0&b<8PO0uzMr|;I~ba*ni|HA8d!AA7ECgqhFIO`Qagr z-hT_Nlwr?qP!4QgyZ+Qvty__rX~eeT;{mpB0IKa^y9ZT9lK|~uuom_nxj*&%?fPbX z4WYHKrV6I$Jc4c0Xy{MevY_2>uPg7{`P%LP>A7u*2 z8%R|_@>;|?FE-mGE=##QoY{0T)0Y|8D^FpsW#(6SgDl!5p`)Q`v^)8az(qwGL?BJ! zzG{tu9j6cihNfI2kVgvWlEf<=bETEGX_>%9H(OoAyR7hz*{VDxcK-sJMIOBzo7OUv zgnv)vin$&k!@L5;Z8e1adYR~*M~29dLXKOQ3~uYrN9jYMFlE%*4-dc$w&iX$^0(>u z%VLKOd>dha(n%?_UU0Qrbs6m**}F`mx07rnLm!&4d1`~$Ovef)8p;Jp+U48nF!y&$rTUoo59w!OGdj`5GF<%2Jx9!$TX<-YSM|& zu-E|f_q-f5Z)QI>PfgoQh{@~iCILK7tgQC*_*zFV&2;2e@G#M>Dqs?{)AacI#(zYQ z4?B$t1_x|6vW0^-k?9hAj*}glnl@8JZ@defJrUg>b|(6QwdhaHo|He#8t}+*6dj#8 z{9&YGzz1Axhga^6r>bq|l|dvxzYfjsqBC__pdPf{fwqMn(36ydDm)2K6BWZ7IEC|T zqB5D~>wz-P_0G-t3azt=Ql6(1e}A_!PD^jIL#{h5STe9L3;9 zzARKCKkaMiaFnOkqM<2wv!N*ycdZfyV#WJ3XonMz(t9JwDLw>tJ;N-yef1PNm#&z| zIoMcTEn&r8u!2EN0|wy4Z-1Xgs7kaJ_YJ1@=czeWZFnCfYHL3RpMF+}N~i2s!S^y= zh*;1;yK^Dm2;lVK+C4Zl7{ealx`-y8|8lg0y_z>(PTU~wg6|XDhYSd9+wDWBt&Np3 zit#hRXJBpNwn&83L_g2i8a&cXssXLK*%Q0MTGoF&#d1R=hdDL;M1Rk#izIw9v()-R z%M6My=+Y8jP=_K4LO6;3+T#u{R}0a*^-nl0cY8bpf;~a{R0ra zg|IpDkvk4(i*CUiUjgReu(Ft7wja0qUCc>BQf`X?K7 z8?R1A*uELhZRPxiT#wHWbjqK z*du_R<@!ZvZ@EcO=IMIq1n1EiNMM=&M}}1VIUzq?coTv2gMTdYy*h?)WCjhkQSF5t zBb6gFEM|b`+;O-BjfUJIP@O~79i+)obP@`;6X)rI{x*qm`dd5wjg2Lwn}g4w5%OOj zR_5TIeMAC{ByrHK_`{UI7MRrz)MHJS9Y@GFY?yc_c&$u}#~V_;-5!k6M4N+~%Jx(j zMi2+kTAW_DwB)a*<)eZG@2& zGlOEA6acOOIYW^cW!=E9&srGAs}LTd;-(2X1l0({bD-bJ3(5?8GpHAi0;nqr&`9yj$`ZkU3Aov*ZIPoG4 zAdFVfz2SRul&dNzZJJXfIWA5YB5;vTKoApd zpn(KcoaP6DI9(heXn~8*Q)=;B6{QS=HiI%KsDIGF;rbK3H^hI!7P*LMF=z)Ue;Vz;Eb>H_07f}q3xh5`){f*b z_anOH_osO# zPJh78Bn~BJn})s2a0&|Bw+bA9ONvPRMQSjKLWvoEyJRLXw+DBx=$kSKw2BTGTI28H0v z2pFBQ?SusSJCHOKBW(NZvn?_gIpcKx5`VvAcg}9Qd}7Y$C&5jQOdP{QO;!) zY;$hurW|ooR7pIp)=A1075H*Rg_@F`g_%;FNKzMpr&vEZNYg8xB&pOB0yuyUMSoG( zbM6hB><&RgM8)Xc!W<81QV7Iq*Lanhk-IEL*plK!;NWm6Lznp~s=@xMAQ7xbWbn2s z`&>OD|L^J%Wo%M)RSrORRGCmgLYWw}!2h=(0VPhz$*nLTb*iAuP^NlK$2{S(O@-ET zAHYyUk}k<_GnlJ0VPlS}9bgvv~e5HwgJ9x#&gb6hu%W@P@5?hC zOy&~ASwy)RN`hijQq;gwlexpA4+_1$s6`Q-lH z5ssAXFXHg8bF7%>Q0d*sd4JBTLW);>KxH6mlmM`9I3I|E5H$Cs2$ho~wYYrBmzgiP z`r*PLrSCbP*Xdw`xla}eg_URlfy=*_!JdaO%3{ch4TNBl>BTqmOpWgV*@^AOBqrr| zJM`pd8lecTpCtt;z=r1}Q8=Z!G|AS9TWkUa#*NPk36UGr8{UtaIe+u$j=2iDV{_=x zG@E)(5+N9Tf{Qb$JJuce6-m!Sk`8r18}!7S_@`1KfbSyy9#EXXtFi&~S%1-KJkPt3=Z)-0&XRbTmCb+NgxLa>bwQ@#Q0zvI_0w!>(j>2vn5q-uT@NOkj zi7ajSc2C`ph*C=LQaQ(dlo>1nSn@w` zLtYhHB1ZWVGHnH+Ron!LD%38iCx#;fSeK!AjP=mWgvO!4w$;Sk|QMC?(Nx5)9GK`wlfj_=4V4**NU*h%#I7hueC*;0R?z7@S{fJO`7gRM3wJWF+J+-l@Qv`L~it z6Zuso0%wzt9gam@{jFR}4Q)Y57jlbAEZ3ozOlg<#6F*5vu77*se)}V%5C(&b(xati z0Bb~r497m`oE1}?6@$;A_bO5{{i$kKKf4UnM}HQV8?B?+xw~S6#g#Wj0JF`G%auE| zCdQmJ=J#U=@7&6GFUv0-WqCs`2dH0SeGT~_bXv*j2K$EkZ?-2`q-2v$8dFO(lfEm3 zfkIKWhkgKJon&D+5qPBqG~ZE&wE!9prv*zy%x8_II@Jg&3v=jyK>tZ4DPw*Yx zID6Zgt^8b>Jy`iAwbX8Q`t!ew6T`X0GRO^}Sik96svRlNxN;&i$zAx2hI8VuIl)k% z6SIp_cE?mL4{mM?^#B{^@*MUm2c4M0%wvQV6c%f)p&vkS8|!|!V(3SDW4!|0<3A#{ zD%odHl;;@uGKvx>;EY?DaFj;Pbr~$}dEBPT+|^#nOsf3Z!bQO{*@>MgDoXlKK}Ab( zLc!R$G6za21dd~T@3;Q{0RdL9khkwe1Y#!xI5sesfi(muf0bF=ZrnH$ea~0)(?SN^ zvP9hj40dD3$t1BShTY>}vAY<^65S@EmO8q`$$tHws-pU4HD>3fNfwJ_vFg-?YHzlj zd9(Mg&KIxF-l=ewxmm2@+2U~)MauO9HA}t7O`>SFC};n2eDCBxi;u57yXx#+5YJL9 zi9=d453`ilf6P_nW?mp`?%k7l6nW0gQ{@k)GCfg#%53u}ZI!o6yFr=vd8hl%F0V@x zWj42``@S1SiJZ=7hLC&NHnit6A4-JcWbQl7YF+6cf1BfoPRCKZYEI^$mb&UrK}j;& ztF^nCj#?i_+p*ke&T@GBBFKqJ;M7`|)C--*rWKO@x97)&(KQLs%Ve|e&C2{Y?@dz+ zSFa12f2z&5L3hH{Kjpp577$|M*e9r^B7LdMY93n@xbV>P)_Em!vUOfnx?a-eM9aHQ zZeN(Ze5wIdw1@2VjYn?_73k-bQ1es`@!v{{Rj}BiGkI`hAO+ zTCb@WIF-)Z8b?EyWDGBHp4k`^$M4vUIx*`9odlN@oRL66S(&PWDl8)%OYJ&u^OdIC z_BfI|tz{voai`Va2GiPY1ep`E-aNcw>LFypR3is;|7QFy(3P<97ntt69G9h%rLZiW% ze-n~)gXok%l(8A+WotwgEjfx{K?fXeH%P4`V`9~CXaMQJrKhy9 z@8BwY>ENIl+j59e0TjP%B$R+iHquoo1CCf|TfG&}ojQ0i2Cx8eu&Sh&SLEwqq?`w_ zq>vwnmg1e|DG{9s{RK-0im9bVEbzxAf4hpav(1zk7S&?8jux{*DFlyT5*O@s+sUk4 zaThm2XRonZ5np1i{3`)LZ;dx0#n9zT7ThWjEt_mu>DCl^B_uvDvC)JGl($4Jp+p1W zYVB4p0y=2ZS{OPlMjEqL;3{S3$Q!~iwzxy5E1LDjI$wMImjGsw>f80*8 zH*xa8TGrCu=KWCqJeooIC@-cEL2Ef(=9?;Jq~M7@v!tvI}tl5q0;d+m^1&ghsfyKnH@MQ30 zs|R5wmJY&nQ+YuCAW|bev$>r(e>j4H^1%CQ8SJ{IwoCG&Xxb9sh3IMMOnU6EQnc7e z?c12$lJ@*;)Shydy4Ed27K6UkH9H{s%|oYKSyT#z9hbNSIp~S`xqGCP*l0b-(EJ#4 zx>N>PCRfhLr>jzWeLHBG;|NJ4(La$k`WNy}x+cO{%}rDlQu2i=7GMKEf6wga6=jhG z^~Qn~g8ALVNqiDN;uoPe=ET`bRkNB*;q2qQCDn^vFQdnyF1DNSTy6*Z*_9$tF3l#p z;O2%X? z8A0F*11;60+>t7jUnqO88|)Y=H=^eX?RcvZD>Ub|-6zrZX%u?Lo2=?c$puEVUl2%7|VXYy^>4E$&TDf<|au3@#SUE_v`NU*_p20XY0fZyLaz9SywQC8=^Wsyhm2~I4G{yy^ z@w}V7N4-JSZR$Dg391);5)sbv)6fvkEBGV^3H<1+^5V(VS_Ty_k*CVsYB6?GU+s%a z4)XH)zH${_-QEUW5uSM;YIbrtI0CG?f`XN1uct8!!V^>(f0I2r8Yd_bBkvAnFZPE7%h5#=YgMhfskS-8@E z8XoeI%lZy=uaHlMR=YFog82(KzHLwgoNu43(}8rZcvJJgok5# z2~#)C4ixii-nNEoo-`8wN)qSRwS-TW@R3_M9xjc%f5fw~b7!d^E!TJ3mX`kI1Y6`) zZ#TQO)9;%V3i0mrn89zvwy;;^+#5DeJJakV6&lWSzuu^F1qh)zVuxyPk#jThL zSuL9a&xEgVS)2eCmz7>2L3c8unu*%GF>`S8s)Ts8$$&?Lj30Y$whc27ERB;LV<$28 zf5u~4YAc9j`44g-qNAT|h3I6wlfD3@F@sTw!x;=b@lLu*!!%0fo-7se1ZGCgbWUK6 zGXMWBmjI&5)L-4sUd~0CEO7FYv*YCwNg^l{jFAA`5h-|TsuD_-i)8U`WGf^MhaZ;Vjb(U`4}vK2e|k&4x8!SMKdMx?!O>#joW2jBK3^ct8n#%PX3l4# z;(wj@PzAq_iFj>U&#R^`H`KA(pczz9cq!CR%_BJ(ov7jf{u<2UC`0#aTIUExFq=isWdu>E+}FBsn~)4`@Aj4{mBICx!7f!DmWCOu9$|><>J-< z0Pby9U6%pT1QdtLb_9pYb_IvZb_R#ab_a*bb_j>cb_s{db_%!3b_=&%0yi*|L0&0; z?Of?|+c=j0?!SVcwq|O|BykYOR?qClahy1DF5Br`)itHqimhqca!X3Od;a?Qz4t&0 zw4~UIy|uGjU1h-__yFP^co#r6wJPdDsk*RDVWTf1t_ovKp;3my6s9pnV{jE!(P&>Z zF5=o1KA>)Vq3YUhH98bZ>lzTTPzHm4G!_Fj9(@cdzCp#RLPbx#Yynt}$5O>oC7=>? zp&eR`MnyNMIK8DcrebTlMHK*>8o)Lvz@W9!m@cNT8%He;a6rk_k;bB@qtyj^I)DS= zLb$~`ZZ*~x#?YLBx^YcF#T8Z?s`$dXMx%m@I#^T)>xfFjxZ#$dgo5A#jf2sDoyDA+ z#uhFRrcG10$h|-?hpVRXh4-~XB@_XVMi$9U)fJUm9R z!8cotN8d)#$bqzoI1Ew#lO z!bEc7qVii!#CTqT5(cDI-hlQ7%t8%xz@VrF(AuEC3bi9aj8wxc4M?wo(bU6IjphR4 zt3dq>31x6~*mA%?g@{=>k~Yw^Mh!jCq)-FCF=9rJB&r4#3P+MuTX58WDaZ>RhLJ%H zt!m(9NIJ1r&{sh`l@4)>#?-1IJV1NYAPA1cS$W8y({wjt**F$z6{yBiPjex0)q!RY z+GtbL3V~c`qj+&ah7Cky>xUnwFjR9-*u&Ca>{{WfyC8k_}z@-MYFTae(KW+W( zZ{Gw4G0Pg0$$^XN;bc62nI2sYPKH?S+~?I`J|9jmp#%8x(Qos|NAtmaSSW7Ts`e(A z^WujeifRwYC@_rS_ORx)5A6k(s@ClVmZ6X70yBUjNEf6Y+PUsQd+1uV7obVqv|wNs z1dT2*@9N#lPyZO6$Q<@BBD7Z;V*i44S#%xbtLfzAXgDuE00L!RYem>cI~`@Dr*BC-c$dvO22X9PZQaPv`Ub z)$Bj3YIs@xI{JBZMRX`9)3fT=YIJ!z{H?q?zxuEFXg(f){`F`+IXNH9=A)CppG^m+ zBT(e;r^DF^{{P2hd^$S)tAav@RvS?=bW~lc+JtImp=;fCt?O;L z3UOP1|4iF|MFZVck{NDg3&x_ETj33MSKA%|8l|8{qOnj zgR@aN9{zkjxSX9=hTGHO*>pI6n_-yk!S?BJJ{XOMr`zawHJFZO zXoV~G*~2S2fq}m}S4=21Y?^86G6c9b)|XzJe@6s&Y0`S(Y`_rZbVO6w{9JFL$!NmxF zh9Kegl@wspn4uvtml2G&D4nOLLmWxyR{lex27RQY8OM7l^D%v7e7HTYBR5oXeuhiJ z*2CeTaLrRM!5O74LQT?*1O+g$7NHU)VMxY3ElbB^BDa7sgI2T%gC>edUsdyx_b46X zOVoGVOK}~2nVNyJPFUKUMWVHa()Y@LKHaP)`XaG~9@f%KEb62|i3h|Lz{J31&ClW$ z+z=P7<`!65N}susQh`J(N?{-|$ed`(ND8xQNHU`05x&2aaZ(u^T7qyqgOQftxKuQ& zpt9nvg)qzCvG8$A*NIi0+VY@^UwSkVwm8 z6&bKJ11^0{%Gfjn2S^%B+P`cUfz%C%5a!e6Ti6u17=ZSu^hftymLi zg(K#ovAJwuRzjrSg5xDB!ol_6G^-h6K_G}gCPW^LG%`2KEV%Tw zW?RJU&Fa^&q6kO^Z4^;A)wO57sMJGRTRQNJSyxI+L=a-yo*lG6SWmdyhD+uR-~GeJ zeWx9>aFLPQK;#YnQ^g6}HY|yWp;tyIKw3C~+!HRPC9sD@3O^<#9~F6jW+d!QlF(>L zZUquM6|COLk3)``)Kp_cxa$rm4b4NOI2XjeVEZ3U6rQu}mIZn&?CA@iP z>~wUGkvhVmWYAlf)K}WE$eYxHn7&RRIk$Z~knS=!@5>>{=axGp$c~WYa49Xx z6#;S`(xFN21vy->Q8i=2aE6Gi1IG^%DNv6{G?8(npNuS}&;h1`C0#Z+~^kni^{$Sp} zE%Qio2X=>>I;oRI%^Gq$7xyh2aB205$;~a$sT}T7#|~Hk8dk~J#wDMM>$tbjO03SM z4Dd#5Dk%Yj(%&o%F1SuN*+}u!vTh#s98uk{K?Gw+m#d z1a*`g)&S0u{h(Zb3I&ZmL3!;spc5xmwQO-U+f){JhVV_~;L5XLbKNybddP{xNHxL9 zQzNGgx*05_JBE&?y?Ta~unGHR9OSxd^-RLKwI>QSu~u3r7%>SZhnJEpC9H3hRHJI2 z;8{RwNu(G~k8o*4QAKu>v8~Ko{0%Q-jyx&ap|(Ihv3XH{vyd3_gE%HiWtO4oWoy`{ zkTP#KhTA3H;Mg#gyqcu+?4wI5S)x>CW9B8VyzB=AEKi;Pd3Nlj%zGtId8rIF&270< znkSn61y&ZR^h_-X4-3UN%OKIhEy@I+#_pbxIg){Csg&p~A-ix>nv{eU-@K~1LzL*9 zhmcX%mC7=IXre;@ao_kCSM(y?l!jYw$z+Dqwd5@zhCmz@YFr>wQMsB-Kg14*yZJX*55E@px^8+}^ z=8S5G7Z@v|+0=_}T)L>cx#xH+<0LIKu<&b9%Q;Q0Isb9|rjrQa4fpF{<$mH)qd78m z2?4%Gz8hzvWP#kJlyUOXyXzxMix)1{Y-SZ3Ksv?cIks=e-S8aC3mkGTq&*RrddHd~ z<)X5GkwoKg|MUdOraCri@&tk8GDc4IXDv?glM*F3o0Cx`$4Oc&@>TR(T*=)W1t z=O0NMQ-+!K3GV?o&SrVV#*z-sDRz8JWMdzHDca*)81=O#vgfBb{li9Mbo|_h$04PdO`t2oGwGu2juS zSz6w~v1I5-b5Us$&n@Tp$YOwFHE*vuRT|Ogo)sA>4fdLXB-#uoJt@1R*YFfbJH;T6 ze=|XKCTKbijFHmvqBc74T$6r_Ob<;Vh?6`wZ&q90-y#9Yz9Ji?ZNWC&?ie1#udM+;pN{ zn2_O#g@)Y(m+q&w&~6fkHvCSTe`q?Tx+J{T-Q&Bo*#gn)wTuKjSmH)j&$!2}J)Y{t z7$MIcbkOM>3CH(F_8o;G*(&m}BmA~#Z;xUG57@s8geN*O1WUb6$qOU?6wMU_|_@bjGx3?6L9Uj+>2C1q$ z5zIRXka(vy@JSNyy2tQc-inKyMUp8|Nf}U^N*zVNu%2Sltt&lSp_ELK7Uw%hut{VO z>=}FV803aLKuQ6U2j}%1f6~g47*Xv4liih2HXrRSWj;446EeJwFQv2y@}31Gp)DSg z=6%IzQqrkOkcJHodBl)bO*lHo+wCmx0+UAO5mG%UKBiE2#7)nQlCqAf*WlP%=EPX8 z_edocML3w%d)(*to>VxJzC<2joybxKFECbdw5ejp#{8}zB=rQJe{xuV?4jRgk3D&| za`aHNwq%Cczm&UT^mJwxT{;`0QpPlyMOIvCEU9zwmcoPNGqGQug=5Pew(bmkR?BRg z6@@4HaIq^{J+p4ys(C(|$G75-%M9b9=K@+v!o%%3tkbQ6~SwAv1^t=jYL#@s`EaGaX&cCsR3d{bF!|J0D)`zd3yRmxs@H-l-Zl z#)Gq2VFhA`Uq#zSR&868yx<)j*%7TC@asFYeDP-6;Ofb6bap^NO{Ly@P@gu#6ikI?ICoDFMaXy&RN$o%3HHTN#pu&p;7lZ1w8dhi3 zsJg5s)#P$mUD5k*%C*Ko+A5iU%Whg!W0Q>2oe)P@Vd_Z5Jc;>1bJBPf8x!L=OD{r^&^pn$?~~62Vurc znycz+I2}z+nG|yn;A_dwBee5j@^sC-S=EFm0tY_Iq*MCP`0N*6XnxfTQD~Ky|&v~HTwdsLa#oZj%HWm!S9QP zIo=GZzM>djo}Ny=d>RkitJ#R~?Xx6q!jfmZe=nZD+h>-nSx_(=;0TtLvb3Ul9MPL| z#M`wTxlVu{TA|aD)$<3t$M23Y=jXISZi8jN$4WG@K_P3ZwxO>#aFL_QmmQ+Isn@P4 zuXF5npeyMuxr6EjbmhD16JAT}EXnC`jH}_*Y&4!+s)}xXhH=4LY5RE>3@NH#lZmEc3y;eRBs+ZMKA04|1I*;~t4_^Mr=(v^W;0qW@uUd6a=!9l1 zI-@=+ZX-JZrwmy0_~hXE+YgM2UoFoqD$3s*Du3$RoT-l48MW9ZiKi-WgE zN5YG6D~x*`)|(DSS=INof_fUi;>r`1K7iqYsa|8pU>K~-kS)YsNC?Uk^AR=+)YW!tZ{Zjq%WilV0 zei}1Wdm(^)l?NUdBbl&-AhXeL6o#BnhktNWN8{6B3Q4}=y9lrp)jyYmlem$=c>Q$m z-98zN!)A?EY8j0W#M~g}^{u>fe70ttkPkQzNl`~1sm^RvZyxV|c=Lc}>~4i&4!?RK z*IoQ!?nRVoz4tXOi7bh!#9B${SkPK2TLh(oeq#Bx$Ru_qr+W4D!;|L+gqFV*8Gmy4 z)nsTr0|bxvLIxjKlc7h1&n*eEsGJtfH7Y>hmj%6xWUAM%f82fho{+npY0BYO`w0u`>fhtEGajUnCqJBcFJnCsuHdAOP1s4DVQ5oOV~uUymcR@8HSnx6ht1QI6J;q!p4F$q$U{WrQwovxdXktkw?J>wxhmgn?rpvyHNL&)y$B zK6uKQt&@J!Y9~$Sb?ZB0MvG)EX1goxK)wJzXT~z)f4dY`X!yS3_?47KpJE%O0{3>_ zAHOUt*S=PE7w~r_7`pnGB>eFCaos6a@_|DP< zR&6}Iz<6!`N{rkXUu2Tn#pn`F*~ORfd~`Mby}BHH8V{<=;Tb3YM1Ru=@n}-$E}@U~ zqHnaF48}d`sYP)5>frU$o##xN+d1^T4y%JxP}0&DNHxE{pHY&s>zVV`OSYrST1qy0 zOt2V|zF$pf()TR_zkhC+p~-Q>f=6_@;8D}8N7Jq4(MFlNgQt%kzkG?wuiwrwO&2TK zd&>Q|@XfGz^==ONA)D;xyx!SA{=k^Un-II6Q~&;?)_NsGs{0^0Ig5?RmFBJQ>Gf%_ zktTls@bR-_%1W;H@~k*{^wHBhIC=Fge7x4-ebV`QWU)T<{C~aEWsmRY)mFdf_4T(A z=o8DEgl_L1JbL?xiL}l;N;I-`<|f>kM&F3sC}l9X0E)B0B-3_+!o2&U+biFjyqw27a$r^XM`vzpejbPQzU^e7$?k%1D=poLZ{pQaM z?~SIjIX#)@b1+C3ylww>bUHtu@i)*J_IUE<5*g7`nvvgpM%T_>T%V9#nvm;FC@d!A z-13CN<`cSh9OC+f+|q<%HlZdbkY1iorwIQA(rT_k3YXFt3=+5C?hEcwDl{=MFf}zW zD=;!AFfb=63NKA>WJF+wCE^^A1gS(6r6P;!Ux|nogLYvLk%4~%SVHR|$^qeLk*`GSCe;Jd zXdBSFNOM6R+Gey)(hSH)YeefH&4L275wzW;g&;;7L)%4K1d7oPqP3HjfHi1`(Ar4X zf>N|8w4J2uKpEO8v>l}7paShQS}W;#P>FU1t%Y<0s6so7ww-h%s7^#y6`?hgZUURp zO3<1}Ye0W3+FG52w6>tii250dBkMgvJr!_nc;nfhdfma(~elPQ5c`wUj zbT6Z0Z!ddeYLAP(T#R?Uyo+PK9E(rAe2VovrS&q`vPDIAH5P2^s~9sqeL-GZ57L0= zrA~hm>7g5mvOwLZB zu80{%tfA3co<{T-&eOUcn|YeuV=m9~;IWivLGc*LV5q`!#oc2nAPJjPhZCCfQo{fJZ|#1$>Slv z0G7ZBx-kjlh-r(4KGb`c!3u9H1oz>e`4k(NHacamreVAqHb6KS!mGP~9j znSI$6aJ?hot6h38-i`Hg*1A~l#kTP%5o=w1h^U5Wil_nX1zJjb0eeq8L9_>GJ?#O_ zTsThD4Yrdm5p{tUvs*uevHUo%v9N!gx$huR8)zpzMzo7)(d^gH^^4u019Y0* zKOb45ht3xBxUI`Yg@3 zhBjsP@NpPH@@EZAb3YAcz$};pN5D}q50)800nrMP+0^R3rqDgTA%9&D;sg z-Wk=Xasi85%$^dn*2QY)P{!w<+@c*;>jC>}oFUcf!+|Q+)*AzMuvmXBuI^T2tEbh@ zwO|XV16x5oU>AzD0jaT@fI3VZ))^Cmi(Yb z(H={#7Afkrpdh9!X#R zMCpj7Zyk#)cce|}xaEJII~3A84v&_X| zMMo_2#Un*WE%WDsqIt_Nemk=KSD#V3VEOOPMV8Gwt~3<1@rI&h%N|=+v|`yyb&5_} z_QsP>IAz(}m&4D#|0Q6xB;fCRkrfPvCubkD1pND4z<*Z)9)5os@aS5&YBvLtw*zv2 z3drUK#MJ>MvjL@V29#e7sQfFSx-OumBA{+EpyBm^Ca$)=70`YzpyU03?%x7>R|EP~ zs0%s(>2%Wz_JVz&59|lpvUKd$Ia>#7orZNP&>46X=wHV;&~b1A=pV);n6iR_`S9M! e*P{OcT~w}Y3T2nWQVbjmHZnIe3MC~)PeuyIm(wQz delta 19174 zcmV)7K*zuKsWIxMF_0z#H8_`XMh7X8Q51iN<=VuSRMJ{0SCR)T333sG0E0v8&96^? z-90$CSSK&Upy%$pKhVE9-S{^@e&=63-`suo>b;3?0=qLd4sY&`H<7(bl5Ceo^me%U zyZ5;`4OPA|f!CJ*$@{XYH(L`YUS1!h@$u4^=UG?mm7Vmwt|}q?y2%2s8RX@|W*dKa zMREA^=AU?S?R;Oh*DNTzlrujG9 zAlfB1kXa9Xf9d{2+XS)q>J^@mcPjgb;eIE?p=`8OH?O2o;vHz3^&_vW<%x3>>*COQeK(2HZ^NvuB=)AC)>>!z$?iw#a~g9RSP9rn%nQWd|u zb!Yd86@q;${4*A1cv zG$hNe^~CqRoVAe}T|XR3RyKMmwO37mWZ8(i6~FgF5Hs>8Gx8T^;7BCGko}fe(1@8@JSPsQ2u{pvGD3oy&US`Evo~)v0NT z8E5beQj|V_Jz_|Tv<9dTIqku*(iz6&L}ZTWiX$( zP>|>Rr(?(}MF-A=vFMKpqe*}8B4L^uh4|rWz^<(DhGILNm^~auh)IOeGe^TDW-7Hx zw_U}~1(dTYPTS|BU0}VbU);7tx)sjLr71TkNV}V}gfHuo4DG0zlk14d2aB5Ym$n?; zFHK-$p#HwiN!WkvyW@B?4H|IP=*j^ReJmI15X&F})E6NlBfr;4-*~?6LNP&)v@~)R z3np_+hf34RVH#Zd2T)4w$&ny&1EEHmxIvhbIm4ru$3os*gc{qk=mc7uU&;d+E$KgG zYg6wR*Qq`_uVyc!;{gQ$3i_-8>ILzb$g=O@h>qJ4OY48V>p&DkWAd@|B7z?$9NZA1 zJV|@F$dgrto=vS&n^t@Unn!t_ac%grQ_WVt(}XOZC4yTdkS5thfy(1;Q};#PcQQ7L zXUSVaIc`P?oWIMgkw2oMHZ+KdlI!7lKIy)(tDEHE_TE=yBl8ZQ#8>_tg! zyvrz0+y;N?E)3ju2$JKbKdYKpsL9$cvV!gsvYaUW40e=+KwPn};(DzMQaVZ@yceSZ z2j7U}`i-)v0frUW)X#J!ssMr{l691Kfv)q0E?+w4ciXQ}vb-(xQTlhFwCFt)4dMdG z&h=2gln_(-U=vEX!%5M`lZU$Fhbf80{AGm^m>_>M%jtSxqtn08NqWtEBthzW&PeD` zKeqW{%Gj{_J4z15cpd~{;(KQ0Y*rWlTy(f8YvZL7=Si#lyhz*w@*38&BBQ|U(lB0} z^o=`dLrWcdZ2pKNi4Uh^y1)sy3Qx7*=Ph#HvtB~Ek{OyH8_xH$OHqEJ?b_zl=I67X zJo0~jB3fRbhWw=B)P1O@yl1JlWufjTPO?aMS+d?q?xTzBG)q?d(i}NCf5atuwW}+_ zjT5=jEdm=^<3%eHaN=FNz6i3XG6PGPi1v!s?9kYq@5ifVxG+YNGG!N>K)FPK8t=+( zKTu|}?gp&COYLZ&@rr>)O9Q>F$ylOca-)CF)r}693;}GrpOVnos87sMl4a2sTq3gr zRdt*);iYwtPW_WOaVlocV#y($CDdd4tB&9!9O9v@^Y%#y+nVu z@OQN1f+ufvhaWT#UW$eo)7GPJ_)IC+Qao?h0%>BYJ|8zUHd|cjoo@KLB5Bv8_A+_R zL*jO?XUS}u1iqJwIChD@RK_2imoIbn?b)htC&y0QTlMW-exCWkHED%}kjVbm<`@w4 z^UNH->WPR+`)c?DYcpmHy4g!Xmpqf(2NHkM(`TVK+GQpj1H8Yw86^@X3v{y)JUgOT z+!w16U1wY^yG&?sPg>NFj;Fztr9_eM-F~?H{PEo{Z|`nDevql*97{vXFhG?Vm9?sg zNEPLSOI^p_{<`e}d23+8Qy2sHN*{C^nPjuNN)oh#^WCcUZr=@uvfS7%uiwZ_V z4Zv=+(g+RBk8mjVq}Uy?Gf4?uMEz=Q;)*dF>rT`td6B+Vj%X^#U3@B0qeL3 zcIIeQGOjN8tSiT;iEp&(BYv9pH{5@~hEn(g$EkP$I9|FUEUnNH4JLfU4_Bu)bmtAJ z;OTmq^|T>DkT>e&HJ5&9J?1g;GCxu(M>@(yDcC8XQ9IP6=@gB*0)Eh0h|rwbW0V6@ z2u?%+8UtQ~CUNq9GpR+CDKl;v4vQ6ov6OoY2BVjxo;e1S3hWk-xM{-a7bt%M{KW#h zq<~$RdRJFlai56aLvc*E<~iV+KVvnpOibjW|J%TYu}Ejv*}hn_P;CmBr)&9mf- z8j(7kRcml5b%LTgQDA`TteixACq52!+|ch6K|EnPxN|MED}NKj;;(~P{dEvIOb0Z# zQpI=#0(UHfJCtDEq1{ODnT8JRv8;1W)DWLJ!Kg&Kk{&83xZ>_* z|AX%c9jR2YZH`-7dP_l0sFE{qv8OSSL00QfL>h9Sc;_rvzESa0tz`AuIvg->jZ#C} zvLZF8zQ7kG4(CLO74n^pV0R>a;>3w|Z9pFjMh>GPE3U0u-sm z1@<3lRtMUl(9DA3As0L#a;yr1Y1BOmBhn#QG}z3w(U>}fR~pQvlGaHG3iTDXt|>Qu z@5i);M;_&QC{As`@|qEkL^?EQ#nW}oTpm-JzN$P78pnSwyzC@Hf6wVaD_xRmp>-K> zL+3hJN9nVPDtz?>xd@CLFN=dvZ4pmaMPUGW{&u^rQe$vQW*4@pTCG(}){mBn1|UtB zc;E?YnU4dGEpx{@UqFq|11g7=-CgcM=Z)RxuaLftjM>eJ0d~EI-D}D!gE;A8F8cDl zA8WJ@Ey8~bUjYp|iS9v`yUK?Pjl~5LwajG3J}Js3PSKdp<$?3c#MR$4{wGOKKwF4p zT2np)V&Vu|+L#rlhu8m#;PLZJp5D1kOQ=j+**jO( zzCM5CXJtOy7?BVY#fb1$s%(U9+&$AO5ST~w{0xz@HH9vUp&{0^a)cI6ijm(*ux9p0 z2~0O`Uo#d=oc2=&CYU&S@`HpDITvB>ozmq9;9s{Y0G>G3&oa@NOcN`AU83k1)j>V3 zKUW7Wg{~`i>%UZ{JS#}R3xb3zaOLgz zQ{gP^8#BTlqcgb5SkftBF|pz{XwW$;PSNEbP9SxK%h2Wn`^C3CBHT;GJAT^(O}~Fo zujVNtpx_znL)1S9b!nqHIp;RVVkW0Y{){B5?m)G|@a5Pcx(i_7El+>L=Y9!+)9lcu*d{g9>zwPjo^r(6ge|Xh(gL15cm*S#7)P#S3E_!-#oa5h}(3Xa0{8SQq_InOGmlZpC9$NlG zg zV_)JNI21-=g(78=+SzgbeEJ0vl%U;IB{fw;pz#L!@pZS7&0&*l{_^JQ$2VVor_82| z%UtD~$LCF!6>*yC%`VB}BFi?9`^`V1^5LJ4e|wY2AouX%d?}<%R7q^aJ!Nb@XeJB|ki*(fe_sFx^3F&zG*2n}AN2vFQ%n@b3HfF%uu%NV?Wv9(I5Ee66ANLzfJd z1`nW$sIJ=j(pD2(D$$>LhtTo_gUYDqq0=}G)%i>$?pj1+bvn0pgl(Bj(YE~cdHEmS ze~T0FdJ0~!YR8@_{>oUF!(yjw``oueu7`ef+|xF2opc0jsk%LPf8sszey45d8fg2l zZKI>X!p7ms&n;grkS2dy;~O8l=C=zT2qpkFS00Tf91xLG4m(lK#y~{LuJIO@S^r0D>r#85o=~zK$V-h~M(YSm)*oqk(8*Y>kc4FN! zau3=l7J%~?0VjV-(``e`(h7Ej+s{(vZr>E5YOwIs;K8so!f`=@U!_{Lrykz}%Qam4 z1o{^CH^JF=g$B`Xq$7L-D_7jOZw6b_EE&yWL6~^c(1_2x+DwCLuO^)c1&amP{+@?} z=FRNK=Ba6$33u{(J4s+4CstN_dVH;;mu5P0DY%&ERuz9R3EF9Te0^i0$A_Ip1%nN? z8`;7^nXc&)e2$YHnwln4bl-RuSbL&-f7qGm3)Z4PHG5M2&}+aY$5C{2=JpRQ6%#(- zU^~2WZ9G+NJC6)*0`%+9{4P3Emj&uU+Z||HsKI@ba!`dQ;c22`cmto=RO_r#@WT2__E_a<$(SoYn8?XCom6c*`~!YP=7 zgbmGB=7Am{2kkM?8~Il5dp-|6KKI8a%wDr06cv98U&aV(mx#n7jk#4SteLoExd8JF z%||Xb#LlcghjE^sZP=}rBd#rdea!>{P^w&dtmG2vXOM$xKwS=UIKfg(Zsf~CCGyk0 zh89P8YAqU?ayJ{AGI7T$Q6N^lPmOjsaVxzylAPj0V%Iax66~v|(7JTQMApH=>Szfo z_JV&E3~CxM04IL?G(uIPwZJ#%+MlQ9RJGxG@KRg*G5GYeN>n;!Tm|3Dc%jRJ3fi3u zVI!c^gKPKTP~aW*_|`=;xpxg77o|qp5=5IyKCA(O`qE|vk7+S$-!q9&L zpj9RSdglN%_n}9HHbMMA8IOF6opk8913F)am}y`M@WR2l>EZ1I&+D&j5H?<|jI@0- zp4-adhFlRqfzOH)kbpZd4trwZ`axw;#;>KQG%Sb|Esz0;;v7<$m<+zk7kebovs}Lj z%`GPh$~+wpL2w?O;R-DC|HzbzKPP|WrwdOalzxywKB{91duC8z8`WMI7^xhYVKD>T z=MLbKZ8QXjKy?mPcaS1`(MibNP8`w&{cRHC^tX2U8w-m|H#?s}Bjmq6jLZR^eZ&hI zN#dYd@rQ2$OJG(zP>(fPc5ESR*zo0@;I%R>h&QBqyFD1Ci6#e_%Jx(jMmB#Aq_sFX zw-ftbZ52pl@L;fP4k+-j2nzV_g9q;M&ID$d1l(hqv~Oyu6GVm%w;o1P%nXWUQUEvt zcp9B3)C@3?GW>7C2g-}=IprHzUW(uwP%G#C^Qs# zcL73B0*KFDYQe3FLWX}qn?Vs2RA>OW{zUH${-3Z#E+SgY))#4M$}~n~ND*d7A~Oj4 zqEKBKQzSGp@f}-1_J(}%k;RR10aE&>(GJWaPh<(8mE*P0==@{tt{fV_uSPd^xvIOO z`&Obc6vWUj1c{XO!f1~2!i8^ixx`@Bz`?L}TlH|jVI`>^iyD7|Kg~070#+sglo)Lq z_R_;CC@kM9a3C%rBEgGP;42CRX87%r`NFbv9b|we2{J&b;tpfGY>oAC;^J49do(gs zG>ZV}kelY;ZHrin^X-Z&+kF1O*Wl zqjw8)JfKM-602R~WokyQvKV1Wk{5x6!>J5i<}0rT>#KxBFdvb@+p6qy`H1|#%SV*4 zN!3+40O6=Ip@f7oF=&DRZ$bi6oDh>+?SRy&f;2;!>h(S538!r;w4VFG4mAYnlKeJ; zxmpt@)|r17!NdeiF#iI_T{`|DFq4Y2CFICkWAkOyY`%;$UqEx(*xZ<}yIGekixbSc zm^4mZZjFpknTZjC0xQG=dUAg3%Me3h>t9Wjvvk=7+M^5sv!w?HzS5hzF3HL$}gv#xUEBCYGQV}rxw{@Ia^H`!kV@ULU6 z80UXb>0Qrxj;cb4S1_P5;5CW^ST`IG#71zMds2kT$&p$hpYmnq3(kHxH%RGw;PW~h zY%uo8BB3x7EhK>adui;s3B4?)tk^&0g#>8ZcJiQez!wUIMc|D;P_cm z@CMlMoWu*KFqbCTI&q0jpwPJRSs|h82K9f2=c8iIJi249g6h~DIyB9uo`Xb4225~q zCUwWU1HU5ac}UVB4`_o(%!z+083On&;_nd!1YV^Lpw9Y}dbo%Mk@-XD`p!)K`Hq9c z7_&$M7_&&`a0!-g@H!IbfI{K&v@QL90MxaSKtHq$;rp0xR*5%;U;jdhyO21mPON{B zXu|iIww972z$;XQDtLnwUl`qB5wULXX81%t2V5DK3r{F_fZwB7uEbefD$fJhK=dPd zmijjLf_>(^<86TZY9Dv=jj2}7$ciJVeNMmtPSugyjV+?jxB}=_B9+MER=h~!<|STq zupI>BB^Kn}Bc^@@f`J8Itp1j_1ZICSPD&-N<}dX4N%YOTAHVtj=l5^l{TMt;iLglJ zapq1__RrB7iCX0~h2~p(J9IFD}zo zHnj4aAW((cCH2H`WCH6l6pyhUn)#x!X|QcIgJO0RjZ2tdfp>v}C)U|W1kZmXoZ;+T z2tUar&Iv{W*oY&O;?O|=x%aXua41aL=h==0a2e}jq?yD>jdwCs$x7Xkj2DUWLJ-J8 zvb2_nZ_hMxRh!@KDh{mI$_wiM=5{m{u}cSLxie+7 z;0`MkXv$o^%noQr;VGHO0Xlyu$lj}E(rS<<`e^}D>fwUMIT!$(@0FlY;zB3*BOWrT zIqDAI$w!J{uz| zt}<~L@Q5;7_7A7^c)?wi5n*t+)OY}sZmFOj6G%%4F5bz&q9eQnt8_6FXOoY>h4q8W zW!iXsOQe)g@{KRh(lItH|xF7#WEBL{HQF=7BOkkC$kmDGH&S5dd zVKMj|dM_m<-Jhy<^}Bz|K!Rk!+-M%n&LtbM!1BtIB7ym4$N9>gTHnSTHRcCo3eVh1 zdoRN;9ff&Am;>}L@xF$D5TaH>-C*EQ0nP>mvy^PpNo8tDXVQ12F;FV1_RtS-Stl77 zjs#w*0nKOBVKso3!>Iu+$MUVnRqlpKqm1<8PS>1aife0DB;J1y0kIj=U)T-|hByv> z;RwG~m|Hy_D55>nRyPug!Mgh);H~+QHM|ll2Q66v{5ofd01CdNYiGEv+04(W*@Kmz zQcLn?r$7J0II*0IGlSd!n)REW#oAp3T33#QCb{#U(Q;1gHz()|bYgc=%I?^z<-yHu zrXFbHoS(y5<)~&8-!L;_1)0T~ljsK!+{U_Jt{D22-dL{)_xO)4Ta^qN6a_j4K8&Ks z37~N+9gb3{IWdE|J&)T|nY-jmnMs;ITev8gCp$5kqO7F<6qL0T6bjbHmvKf1D1Vh$S#R4&5`NFG(9?nfESmQbV6f{$ z$=b0ehOKe1*j)^0vMn`2kqQqxo?pLTbro$%rZ77%>SlFS*L}2{$#UXMK7Tk}d^rB( z1(VcHBQKgP9w%Yw*{<(R5+}6dFq|xk$-gbvnf`Zi`N0vY<4=AxNk9?>L^2D~#L&#V z(0@)HpK0#x=`0K#>t=p?cQZd*T-?v;l09~`^Fip@aVpHd>25Wh`Kf1hE5#dK<&DZ( zT`l=Ln|g`$Gz~*5(`ELcOWhIVg-`28Udyam$^}K%Wv%L3A+HJ&Wik`)u4@Nj%-jvn z0CF#zn(llyObL)rXRcMRH>LXd%8uZ~j(^s5J)HqssIol*Bu?qBMsCwBtu9#0H)}LA zayYvpFvr-pDy0hA1=eHTFi8E!aid(TiUjERu-!N^+c$Knz*yvV-8qfWnar3>^*G6!wq73omKg$9ZXII;_@Ky zx15AXiDSmmw_H0o8A|DiuW`8qO3|p&i zC8AN*`%WZ^5;bCQo8Ba`A*vfe*-Pg!QFfEt3*IL;U|LpFUOXR0U)Eek03`pYvfvM6 zS*(qE6VD!Fycmj?j~ZB(yvs_;^C0xB2lgU^ANz)Co!}Ix%oO>#kq!`D@rpX&4kX)# zlhb-k42fv=T1&*npj3W`K!4)&>r~V7DXSQ7#SFtu#)wroj|^7!eTqf#7ddZ4ioVU3 zMsP78R^zgMts0%1-~z}njHQwgfO;02CE!ZHSdCouu9yQhZiT*8Hh0gR&I>HxN9#=( z>sCF&XU6W^?fR{42hJ|B=h{Mw*iC)U@*FQTTi?3tq@r`VZ##_!n18qd8e>)%nNyNE zaYHEF<^mAnjWHKt-%nBe;2IOYgnV)S(P%jIUdIk1nKFDQb)z7NW84yf?6=!1RyvBe zIa-7QZBdg5kjLChL$7RHG8dYNp68HAT|q^i$I($~IRN zLNK98n{}42bfx&CG$hbsupVIp*1qkk*J2y#)Gn$L_wgjJi< zct8=$-$oE~KvkRk(b9K}5*9jlRI-<&(IJQ5uy5h7(7M-6msQGebVBP~ zz~q1`la`nDaypmOZyN2~1fmld;Rv`e!jRKn%DNGXJ2^a6&GKlhs{IOtz<13pwBBK^ zS{-qVzJC?i7t|I3T?^<&=Y2aC5$H~~8wZ2`lo`_SVkqqqSEC`hX|Po`hy@bM8E#Iu z1q!R&o*93k@9f+PVoBp9hIi@1jd?djWYbZm6 z5LkEAzH`MNTplofo%VREA~YF%O+AUXS664dMt^699%w$Hv8m8|5JuV@LN?m=oTT#x&JBh=|iH{GDOC$+L z8OIn2U^Xm;IU1XU63<4n__JV3T|14_y?{0~#>Vm_bq;n~+~oZkGlnBR)k&~3tCSBJZ= zXMdq-U#ai-8>Th&)-mIFAlobci^YeZpIS0Qios47{0A4dYDv8ilNAjei1rhP7k{e5 zUu5PV8GOOfvjf979#OijIU$4aP%JsNAHAzcYt*-XZKCKO-AlIAb}4EM(ZpmsnwF#w z2hCHES%CmAehe=sI%I5EJ;%9tNC9f#nrl1r2wx5^|Y+<78;<#v_U7BpzuW>FHoJ=i;Cvtd0 zic$!S&I}$Tpg)qgW`+>|mK>ZJFwnjgZf@&7nrFDzyfC!7W7Wb7Y$tlH7Vc=vbZp%6 z)*MXn!6Y#rJoCke{{etH;-H7Jcm#*Ccm;>Dcm{{Ecn62FcnF8GcnOEHcnY_&cniy2 z12#D^mvKf1D1Yr->u(!5lK<|%qCXvWv9P-NCJ(cF7sqkp#Ch4yQId#kQo_iha1f!vHqLW*6D42kS+4yGB(_rBqW{SJ@P*gsaL}Q)!g3GL>`GjjU#|3*nZUwB@X=jL{8RTxGnAsQAiiLzPfj?=&j7Xre`huuiBr>VWw;K?w!HMH&aA zdy6?cXMZan2~+1PpSTwY=5gg*s6uEwDzS=qJSwS*2|b)|s$@KF0h44RRWy;;5YY{4 zN~s$rG}q*u{s>S3uy za{=*Hq<)5kGWsTNd0?Po!mK<=8)({~h8}2Ar~%)UFe6VA)qo0>CrN57IO-MT1rNi> zpnrx|4e&A~omeaAt010A$FxObYUKzI&>l4if+ulS0W#<{-A!0Fo`qUPsUZxyK;}aaQ?5YZo(8oqzl=sd zZT;Z^&XH>U>L*gVa@9h=NDL*TF)=A3`5Enm;n?)z9985pX(m9hi=sT0yJsd z91P5YpwR{9UB7$z=^uj=nZy1?g7(To>|c;BOTLACH5s2A4QACxptZYK)yIS1W`9*i zeo<6aeRcNvYI63e`q-jh?+(5WPX>pNcYdleWsiTq8W4TX23w%p{n}SnJx2^hc zFdcuHoD8O{cev^K;B?sE8UI#&BqBjR!7X1!t9mj0gxU2c7y`;Q2>64`%P~g$$f}l5 zAXO{B`C7;PE7T-CEBEuS74o&RAb&IFakuK7@#J(c;i;&f>L>Mn{qQ3xeDxDe9`yXE ztnLM9Mm&e*$}~-{VWk~=HX6Vkefc!QnDxQ%^5^=&4?pk|>IWyY;rOyXs^1*$)Bm5& zXS1v6f7bQjviIxo=iwF6p*NnK)xXxm%hSPcy{q%9|C$YFqrqQ~X5*9d{(p2fJo)?C zq<=aDMgD#|n4aMGKgOfe;ptx$6gsqu4dKr~>-#WsrUz41Ijn`K)B-V03&{OHK&(kU zGTch(UPFdOc_8|K8L{Ykuni_+t-u0l{3g==iSYXlNjx1786DO1RAXX2wNSQxyV1=y ze1){FzkjClRRc?eImTAFjDN$k(q>b|Ohi%iD5h1 z=;HR>#m1PX=aeBQ08=OP8G9+}f6qB9V|s>+L<~I-&t*<$ov^5*-&`gxAmm>zc)I^R z`@MfQ?2QILpZ71P=k=&RosPG^PPZq+)3d?$WNbp9|l%M^58Sa5i8qxNtL^i%?1KW;hdi6!hE#XQJ)h0$uoq zJ@Pzngj1^L;?0lMv481NT=qZZ9u_{_HDA$_gE4pY5V^qHm|t+*bqjCqqf4rXVKys& z+WH8;y@!RSKyo+(mhRl4lyS#ZhT})=dVi+9^g7|@9PQ9fcNq<7 zK}d%or7T+nrfe_!O1t*DM_M5PVVPP^7 zUSE#_Y#K8<5`S|U!32xa2YNchk%VsL4-z%#BPGo^-9uT984}~e?PVRgp&sXFxFl=? z91aTC0`(G{QR*r-B;AqRyUJQP-jswP84t89J&%dp0>*^vN{cXRqKNcW4KI1PFj$3w zc|dECdMU1>FHDDi-}0+9nwUW%l5xKO+oHP5@~s? zA_JCZz<*_ENEsVPfS?ZJ;8$XmF)_Ttq%2xEbyu`@dUCr?%m-`pOf9@?M<{@L!YEikAmgos0sz@z{sZT*0_(V7b zO=~G(o# zvn^uw7WL~`Q5IGPb&9B)>pC!BRPG_Ib2{*hSyxI+f){A!JzHpjuz_&5j!WhZ-~GeJ zeWx9>77~SDB&;o0ys&MD^-g{qa?GTroDtz}TA(~M50UG$=pkvmhsDY^U&Do zXdfeW3OP&%Cij(gEOM7y5Ysmqqy!fb@+_!~QX~0J;`J&1kIu7C;9FY^Kk)|3E%w-= zPn@?{@+V8now6B3(pj39T?lz-SuY_=qkmdD2Kg>?^S&ICa&DzVg6s%M4wurBT@fJH zA{||JFUaA7jcOPZhBHKD9XWoGNP&7JqKS+n{bXb*g(m2iN6)TL!;;G0O4;Oj@RAJ6 z^$w9RaONA=g2W5O-9iSeZpj_nX)Pe-KJC0&_$7AEs`ic`AoUDOPbPom59a;bGJlUW zcVu_iHCdf3YLqSk!P@(l4Y;%h#pLD|=u{4OsbdQ)0FG5MwsFa);yUgvv=XaxDI>fQ zn@eqJk~-Wh4LBZju=;i?@$tssweE z9M%BNlKr4u3I&ZJLj~O4x+`G7fV6wR$1p+}aU^hFB{t6pWY*Q^HG0mJ-%CQOGOKdXt^N zn6Z(1F0Cl4#BMUSm3fQb@iOMflcF7Ji_{Z)7Yz%EAwP&?qFfdknohQceSZom%XVY9 zUE&Rn4Rgt>Ny@-Jx|EV7%4IQTS@Oy&{(*qysq;@@$6m^^SN4>b%Gl7{mP@61qWLGX zvdCp%YC(8dD85-ni56~AX81gI`;5Yoj7&?VL~jY%g`3i(B&_)6Rl^;UMDIL=jJmE= zmO(QWx{v$DPg>E7bW(YX{Doatysl==dWmest>Pgsl*DKAovbeFQ6my|JY zna@kQ7K{U{rHrfRWUprrf%XR2|4$jxVk39HPGYdE0PQ^0BOo-abbsduaFWd%H6LDJ ztca#-7Tvh?Nw;$^@mR)5T4-S5*P>Q(np#W#T05yBhp*TTyE%q6EeGWHn(zDK?r zXQE_*+@+Lp$};%tBTGvVE;Vds6&pZ0#T7WVcjRsaj^zaoITzBNh)aWKO_6g^#Ym!Y zxPN|vWK%sGHF<(Sa(|f;r}~Q)ulPxs5}eH`s8Zr2Ef)DI`d?hh-9=m3hgvnS`KVYU z@VrX8@OYZ@Op^IBq6s4BH)+e8f>ABm3`+s8)kG)nBj<%#UnI*`Msp08=oTcR*97~3d zG#8a7^W1Wdk2)*{IM(p?npZs|8a=QgBc;J!Qe$J!(!BApBkm)YYB8-IG9L;7j*7WJF%G%uUtzU#K~8-c|oy3 zO)UT7tZ?aN9IshD&7B|j~} zw{If=%74SgCYzz@I50-a%Zu9Rz;i?TEipYbg&rulRlFS_ITsW2cDdpNt(;3SY%@qOt?aOIKufAK z@z#56D32KgCr&3yMj>TOOw{E4n5f%0NOow5B6I+n;OVaKa9YX*RU~Gkj1L_;Pk9HN z#J*ul6Q6CQnOgECnU6;Naiubx=OCB7%{cd_css;3hUHK^D@E?S=36XlE|)5RD7?P-uDV$DS~AXopl%qU?a+cTfHQQSQNx*6mrvvc40z+Cl(rZ z6MVj(+v0qaIL^cGynl(NbE-?iYu!G+O`9zcyh(zRO#2iL*#DB`PTcYICWR=oi*gOuBVtU@Mf9DbnIY>j=6`_Q;-bAdf+A z$OEJlAbD_EFMlDe9ElOtE-=Mi31!RC?oyU>lQJX2+xSvSiy$9ZKr-6mA!*)MOeQCt zT!u7kaL6Nuv}(f9d){tmc^8;GGLMk!QSmW_rX_BEZjzLBQk@3R*0LnVa=k+;u_(g9 zqTb;?w|At%lk{cs2Jd^6{DvMv*`2L5S25g*(|c+%412LgSQ+WWS@!s@**5t_ONwp;EP&e+pH)&$%l(w zDe8rF<5tb{(LBBte_Upm7CjfxQX0mLPtp*LWPgdbHl&P64gpXpc@-^pDQ9u*Kd=>* zCzEGy@_&IcHLlQZy}Fxqr38Y(^ue)r3>hvxT19WcDO;a!pdH z^AIp}f<`v9KNW8j&>8DpymB)cUd_f6IdlD@e}Ow6UhKa)eEOG%&v)Lb1~*3ivuR}o zVuxQv+eTJxTa&!t6&=|Ttsn5)JG6Z9W?TR2$zXVPKBHV5j*82hJ=M0NoBOl=Xn6AA z@_%eJs4)D|Y;f@-y@-mJ@^U9EHjHuJpU_F|KjAfpS9QO}iv$<_`m`R@XZ5hYtjG2E za!_B<`){Mc=b6+d)W4q8(|S7iI=HN7_3YPi{iXirU^4!*V6cbR5_ngT-k0mVW3iqh z`)^;oJbc0wiE9XwGC{O)_d^h^*AnDmy?={0Kc0guhxLy{8Yau<`W%EAF=?*qtHESA zK4ntOK!C4Z?9C?FKYjXP=g}@>pVnX>=i-p=346TZyAFG9tUnLG4(iJxM1aHqT<#+1 zHbL<4!|vhnGe+=uf1Swdmcz7_;_Ub0xJ{Ctv2503*#}2Qb1u%h$c0Uidwlr%aDVSH zkbAUtHRB}dV!m5O4OPHG2{!@Fwd4)8C7ESX@b{4Xt{nGp2-ET*!k7Z zgBji2>0TCh8G7g1HQjIZz)pFAJ7GX&?@xm)9G+}HR}JqXqe&6koqf% z!R6`c_{*o!V16|n624uQq)k}zY=8H~^LP8qk~IqoW&<3-vQm~-REHyabB+YNmLt~* z&_OG*Cz?DklRCN?N!O*J-l^#(3VH2JbcRJ&&FnhLtaZU?%O z-jX}0UqDyBt3TnjwAPZG4o0{dTuq0g@ujNi)@K+Oyv2rhktX%ec){)B(|>8dzJN-- zBEgvr`*=6?^XK`pgq0bdVW`xQdc2G&pMw0jB=xA?;C!cOG=}mC=#IJ8?+8x$GqI`%~S-Z5Z8talSz6Y(S5+mcej32@70f434cF>$UcYAlG{px z`v%VITPWoZaCSbG2ESw7;oY9#4bsaZ01yulmbj0%kO#kf>5rheKhxi^=0CWGpV5YI zX+WQ!*XO@qog)|+*8d=-&iY(DM+qTT2N9VoSEDb}`j`5bFXP$Z^wWr;<`)9US7qRF zF_Z~Q2r?c1Mq$YLWPboRbvQa51-tCjYICN{Y(wx!w zK+Fwd-rUMN$7gHS33;CbksNh&k!sCG{pRuhhc^#s#{O0qmhh_=a?{2i=3YdZ)_Yge zlE{*nO01QIjsNX&K3bvtFXT2+Ywt_o50R{EvLadZ^SjjAPXqFP=&eD(T0c|3<}EU0NMsPT8P zptp%smJQ0&Y*Lo#{=R>u{7TDFn2h1IReS^azhNnUS+)e5vuCq7a_7O*mq%dF(ZL%2 z_|_wJ&7C}))M`e3BZG)N&oe`QER^QiGI5tPRV6umvr(4r-RpygPhT)Oj*r*oI>=NQ ze$#-ydQcEVsSZKta-kI{8@HMue!Vf=gM7YXsqD(*;JQ~bT4;aHk{r_!IXFv1*~mM1 z^7`$wCrp&1btIV!NlN4gru8yHm$zBNacx#>2kW)Z_!Gjwv5VP8S-WTNj~*X9Wz5z| zznN<%P3Lu+J7Y$RWG!a9EA2qJ06rJSvf%%GDXh@&UB&S$DUB}0HcAEV?YuvJb<7l7 zpV|*|mC=wm#JhhJ4ae^bs=2sYr$5elnq_VNeu;vk)tY)-Qty{9gDHQ~sLg{Q7=DNy@Gl&RZwhjxK8{>GYbk0uJSE zq;?+cKYsg$=e{lrgye0aPmF3u2M6_ih{1Xz)8%Uyu|>$WyZ7w%qXVAx?fmObhtbL-LcYk9O$rtaYBqsK2_V)E;^bBt?aC3{b~A0NLN z7O&pTAwOi3-JI7u`^O&`vvd<;*K_LMpVV5fgh+KCWG82_5xLU5bv?Z<4K~um?;k#X zc1&5x^EaxMB!n(9$~*u`XxJKcT*vf@UtYNtOP@Hh9C&VF*&#olp52=nQ*2eshV8=qb%8Y(ArFXD_Z#$SzIDcP11U6Y_p}LUHp6 zT{{kOeL{X|LaCULD+#2R=hG^}e*sMvu56dV4G9ys?C=ZjQ3o?IFfukXHJ9N+3>*hD zF)%VVGc~saLk!#sLo_%-Ff}tpHbz51H8e#uF*h(qL^3!zGDAf$HZnp&Gd>_ZG&n*q zH8VvvMnge0G(|KqH!wy-GB`OhLq#w)GD1Q#mm^LLIe$rPRaF?r@Ar7`wXd(er7eX5 zW#~i;Whj(RbROtHX(=6mwzRZ#0WKtQAuKTJLKhmhCL|^rV`3mK#Es&K10hj}1BQfP zjESB^U6>e+LySg}dVYU4zkJ{S-goc0=lg%>oEJq=w3>(#@fwf>sYDc|B8zrhh=}Hb zcIgn2fqw*;M{6Tm1H#WDzX`3CR1Zj_ZANP$%>lV++tHdyGawIbCt4F}7UZK1q8%hH z05RGyS|e#8Sc`T9?Eq;JC`LPqwx6^FtV0_|YalHJ>(Q3b_K|J?WoXN2dr3EfO=xG) z_KDk9H2q9352G@-6|I_71K*3Lxz{W8(G6$=XqBXzumfnCdkt5UeGpq1 z+ICVEsu^u9nm$!UZbj4gZY7mLR!4#MVGvEX`2~JxS70TJCz%0Wp2Ktynkk5CV<*Lk;3-Cc|5iEgau!0YG zl&3X3t>I}1uZExvyxIWsdzl~0ds!Z%dl?;jd)XUPdtB`0V!Z3+T^#G>SbXZ`Q>^DH zt(UnLEGm4Uv0z(Y&Y0=xOY+)AkOn+2b$^CP58XnP1?p}-+EYBg4UP5Sk_LnsJTCFr z$KxxH;XF?CYQ$P@)(6%xe;dF$ixx9h}<7AuA(2OjHA~ zV?0cxrBCl2BdP-I8u2kAEtchGAJj)?AMOnJs5ao^8od|q#(FtxU99(F+jxkGwJttN zR6#UOR0}$QmeO{>-V=`!9RgZU+dw@RP7$?&-K6tGEntt??OS0i_l7kVmVejseV~E+ zju7n!2S`s6H4@F4eRoa2I0%|Rv)P~LBTE!r)jV&s#1k(n+5{@NWytKu7HOM7IoJZ! zrfB7wPWf*HXrvm(9MA{O0!@iNOEa#ajhp@Vdl*6T3k^+kKLL(|NiYSb!3;P778pT3 z(IS!Dd4@=4T_Tcqmx*NMvwuW#$%-Y;{iH7v3?u-icC?U7fwU#(@9Gwe?$!g`?l8K; z@@^ei3yJ{Fb|sc9Nb6^~*mYTQ>$gg~Ejeh4aHs198Y!l6m9}Q}x~}$tet>Nq#&p^n zajc`X4l6tDp?KU;eTT!G_F5d~b^VY=qJ)jZbM28O?PIQ>TVShr`46vEQdJa3f z6|ii{$!FBCnyX<;HfbqvoKRv#RNqk}NA(>gb5!3^K}Yo+1$ASVoOm=Pw1%VRj+#4a z?x?R*Eis5weKCzw%`uY0R}P!G)0VwIs!`>5vK*TTJqu@jlRv2Z`LR}WXXFairOvt+oYlnOBLKz)M=@PX+>R@8ht0S z)aOqt?X}d;bBg*boq9o0zoqNak)_{wSLvXoKbem#XP`;xh=1j*Ooa6QKau4QeiiWk zFgj*6f15$Saavuid7Y7s$1{7ZoD1AGi?1zBzvVd**0hPT0 zH5UWwxVrnffQH$C#w!8M4+7e52Xx%fDO?92oo?De2j~P{pc`n*(y?3TY#pq18rG>m zXW${Ae;p%0$H7sce;8w6-12*lg!hg<8T}8ZH?2GhWtY!X3>*tJH#9j4B_%~qMhcZz B%H;q6 diff --git a/report/main.tex b/report/main.tex index 0f346bb..4f95fb0 100644 --- a/report/main.tex +++ b/report/main.tex @@ -46,7 +46,7 @@ \begin{abstract} - The basis for many data-driven methods is the optimization of numerical models to describe relations in data. With the increasing availability of large datasets and the growing complexity of models, there is an increasing need for efficient optimization techniques that can handle the computational challenges associated with big data. Furthermore, a good optimization metric can be crucial to find the best possible model to describe the data. In this paper different methods of such optimization are studied in the context of large datasets and big data with possibly noise-affected input data. To render the best possible fits a multitude of cost functions, minimization algorithms and other approaches to reduce the computation associated with the optimization process are studied and evaluated. Reducing the computational cost is also important to improve the economical and ecological footprint of training large models on big data, as the spread of datadriven methods in all areas of life is increasing rapidly. + The basis for many data-driven methods is the optimization of numerical models to describe relations in data. With the increasing availability of large datasets and the growing complexity of models, there is an increasing need for efficient optimization techniques that can handle the computational challenges associated with big data. Furthermore, a good optimization metric can be crucial to find the best possible model to describe the data. In this paper different methods of such optimization are studied in the context of large datasets and big data with possibly noise-affected input data. To render the best possible fits a multitude of cost functions, minimization algorithms and other approaches to reduce the computation associated with the optimization process are studied and evaluated. Reducing the computational cost is also important to improve the economical and ecological footprint of training large models on big data, as the spread of data-driven methods in all areas of life is increasing rapidly. Using optimal optimization techniques the convergence can be improved by up to one order of magnitude. \end{abstract}