### Create a StudySoup account

#### Be part of our community, it's free to join!

Already have a StudySoup account? Login here

# Advanced Topics in Computer Graphics CMPS 290

UCSC

GPA 4.0

### View Full Document

## 40

## 0

## Popular in Course

## Popular in ComputerScienence

This 39 page Class Notes was uploaded by Dr. Elyssa Ratke on Monday September 7, 2015. The Class Notes belongs to CMPS 290 at University of California - Santa Cruz taught by Staff in Fall. Since its upload, it has received 40 views. For similar materials see /class/182267/cmps-290-university-of-california-santa-cruz in ComputerScienence at University of California - Santa Cruz.

## Reviews for Advanced Topics in Computer Graphics

### What is Karma?

#### Karma is the currency of StudySoup.

#### You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!

Date Created: 09/07/15

L ELTMRE 3 quotUhEAR RE RE waU ONLle M g INlTIALQE Qquot Pei 2312 135 sscecwg INQTAMCE xt Em 7 gt 5 2 SET LASEL 3g 39 mama M99 91 ith MQDATE Wt To t GOAL THE To TAL Ler 0F 7m mum Mg 9H0w0 Mo 35 MoCH LARGE mm 7m 39 To TAL 1099 0F MI MERLIN5 ALC V S 3g jlyiziijlc nggiTg T 1 gt w w x m 2 Z Wm3936 wk 2 mam ti 39 I W 3911 Ww p7 v V ampL NE Au Caerpuwg if m WWWTS W TM venues I 4 5 CEIUERAL L099 Lat Wt t LINEAR REMES QW V 39 7 39 v 39 quot 1 T t39 LQJQQ 7quot Me GD 2 39 LABEL Pk munw g a m ARE 16 wan mot 32Mqu i w L 1mm gQuAREb Mu Bt TANCE wag We V1 L2 520t139yt Xttt fquot WJC 47 L15 W 39Kt Mt E U lUJCH UM W r xm iift W 3911 L3 39U39Xt z g mPUUT I w gtf f I A Ewamw WNWWMQQB Rep mrn RELEMTMP E V I 39 I 35 r 14 3th b wt39 3917 L gt 5 Wt Kt kt 39 t 392 294 ampvt 91 Wvgtfxt memgmma I quotW Lb wt Am CatEm CSF PAW EVAWPLE u Gt 39 M Now w ER03 UECToR W K 1 1 L a 2 I ham 9 I Xa I 5 I w L w 2 t yt 774 mg Liwgt 3 2w w L438quot 1 Q I Watta x U Q g 14 w 1 i3 2quot a m w mavxtwgytmg aw Mme Md 6 e SlUCE EWtru i 3 i ml 6 1 5 3 5 Mmmmmimz f 5 I II aai q 5 g Z Wt a ml 1X1 giglt39tn 3 J I dtuii 2 W w Wt39Xt Ei t 33411 39 Wm 6 A m 39 M39c w xx Z Wm 6 39 t i in quot t E x p a awn4mg 3M Em ALE 1302 GENEAL L093 L515 WW 2 I 17 Ltjtgtfli39x LEX g 39 MAJ Q 39 V39 Wu 39 39 MMMLze I If 5 b w tore HINGELDQS E Q EEiia im 0 wngw i quot P205 UELCT EEC MEC NQRMALQEE Km CM WSX CGMBA70A2 0F L9E 0F EypERTg E E i g EEECGMES EXPEQ 5933975 f If i Xi i 39 iii a g 1 ii i g E e ng Xponentiated gradient OwTOQET NeeA uwE WEIGHTB Algorithm EG Us39s n Parameters i L a loss function frOm R X R to 000 39U the total39weight of the weight vectors 5 and s a pair of start vectors in 0 1N with sf sf 1 and 7 a learning rate in 0 oo 39 I 39 Initialization Before the rst trial set wi39 U 8 and w U S Prediction Upon receiving the tth instance xt give the prediction 37t Wf W39xi Update Upon receiving the tth outcome yt update the weights according to the rules 39 1911 wit15 where f 72 7quot ll 39 111 Jr 39 39 U o N 1 p 38 ijl wijrtj wt jrt j w 1 I U c N ti I a 39 i1 harm wt irt 1 exp qLQt 011133 310 1 39 exp nLgtytUat 311 Figure 3 Exponentiated gradient algorithm 39with positive and negative weightsI 57 5 77 CAM BE REWRMLATE EH ii A at 0146 1 9WM Meant490 11 Liyt Qtl Cosh r quot 39 39 h I39 i r 1a w PARAMETER 6 IR 4 M mm i gt A l TARGET I O O 0 x in 11An k1 4 o 01 1 TO M L L055 an 39 5quot wirri 39 d 3 ARGET I 10Oquot O I 11quotn n100 1 xin D Q CO00OCOO O CCCIC000 39 C00 DC IOlCCOCO QOOOUDOUOOOOCOOO39COOOOOO OCOCUOCOOOIDOOOQOOOOU0 O39b100 v E 10039 90 8039 C CD P39 D 550 5 a VLQ 1O 250 2 150 100 50 q n I NCO9 14P Q 3 x E 332 xu 3158 300 moo o 00O QOOQQOIOO II OOIOODOOOOOJIOOOEOOOOOIOOOOOOOOOHOIOOIOOOOOOCOOOI o mo doc mo moo mmo moo 10 15 25 3039 50 00r39 OCOO0 OCOOOOOOOOOOCO 0 E 01 O l C 60r X 39n 11 n k1 n1000 O2 O o1 01 02 O3 1O 30 40 60 4 w 39l 39 9 A J39 n HI quot 3 39 a w q 51 b air 2 39 We ef g w i39 u on 39f a Y I V Wk 3 39I I 3 39 a 39V W 394 39b y al 39 I 1quot My Hw w I 1 i quotVV quot 1 l2quot a g 5 Vd s swam vquot d 33 z i mquot V o 3 a r 39 i 0 0quot SNquot 3 9 39 0 39OO39 395 391 I x 04 30 XI 1 weights 0 I n 11 11 k1 quot1 00 A 939 3 A 8 cm ov cm A ow 2quot 3 quot5 l 5 x 65 E f 90 39 LARGE Minoan VECTOR 5 IN t 3 g x 0 HAOAMARD MATRICES quot lcquot 391quot H 39 q 1 H g 3 11 H Hm H w 1 n n ft 4 l4e ALL Rows Aral ORTHOGOVAL 5 I Use RowsAs INSTANCES I i j quotI DON oww OQF 0v 8 00 cm 0m 8 ON 0 a u q u u a O I 009 VOOOODOOIOOCOIOO CCOOC O00 OI 000000900050OO0000OIOO39OIOO OOOOOOOIOIO0000000000OI000000000000000 IO 1 om 1 cm 225 i owv l w L oi nx mui uasaum oo V mwmi 39 10060 X in 4 s rms co 39 u o g o o s a o o a o a n c o o a o o o p o u o u I O quot 239 h 0 x J I 1 I n l I O 39 39 50 100 150 200 250 300 39 x in 11quotn n20 k6 o so 100 150 200 250 300 BB x m tw n 15 r39 quot39quotquot o o u u c o v o 39 o a n o o o o u o u o o a y oo 9 9 u 0 quot 0quot o o 0039 o O 150 200 250 300 02 018 010 014 N 012 01 008 006 004 quot 002 39 x in a p m 9500 no o oo o 0 so 9 o 0 oo00 0 0 o 0 GD 250 39 300 ll quot2 i quotin 3 39X139 IXNI 39f quot2quot 0 1 x 39x39 1 39 x I39X39 rum 4 quot3quot 5 quot 7 quot39 39 P IJORH 2 NORM ARE 0 4quot 5 a f zNoan cs sewnun quot f 7 comma 9 lNORH A95 0quot 39 397 L254 u normx21 nonn arget11 n20 k10 0 u 00 0 135 Hu T 39quot 739 13 024 39 39 015 DJ 105 g 39 j 139 J 100 5 150 200 250 300 0415 m I normx21 nonn argel11 39 o25 quot C o c Q 0 02 16 12 005 of 5039 39 150 200 39 250 39 300 39WH v EGRNX PROWCTC OF outM Moet4 V r 2 t0 L A MotU A 4 g I N f xi 6 quot5112 r7 IIJFtllglW quotaquot k quotwequot 39 L60 5 s a 1513151 kN Lest 53 19 La 5 kllgN I LINEAR 39VERSuS LOGARITW Ic 619mm m Al 39 THE GENERALBOu NbS VseauENces S wer uxtu2 5 X2 39 Bolu m 1 rm L 33alngsu2bux quot5quot25 2 39w 39 VSEQRENCESS Mm xtuw xm INF La SHcUMLats bl X04193 39 gtquotquotquot quot 39l 4 ufx w s ix x QK E x mm m kl I h l I I I x x 20 Vars degrees oz1771 k2 45 4o 35 30 25 f 39 2 w R CAE 39 39 o 20 J 39 J 4 15 g uni n cuno11111 1it39ll4 e 10 quot 56 39 5 quot90 1o 20 3930 43 50 61039 71 3390 90 16039 2 X1X 2X13 5X2 X3 X20 700 600 500 39 300 200 an N 20 V86 degrees E1771 k2 v wc Loss mo T FOR6b I E Gt HAS 60 we Lesa A 3 News 7 MENTAw 0N5 WEL DLT39IOM I in MM mseg THE 6 upon pm Examso E EFF a 1 I 0 N Z I v 9 39 WWWQ m MawLI swmwww 1 39 fj I j ExMam quot j 39 4 k 4 Y p I 3 7 quotLhw wxm 7 U quotHzW Hag quot slg v XixKH I I 2 EggWigs 1 39 I dux WW a w nwmw omrm EMMMARV l MOTWA HDN 0 011 le MPD rTES lame me WM l 17 LC w W p AQAME T la D l V E a G ENCE 7702 TRADEgtOFF PARH uETEM 39 Families of update algorithms parameter name of update divergence family algs llw thg Grad Widrow Hoff LM Desc Lin Least Squ Backprop Perceptron Alg kernel based algs anzlwilnzz fi Exponentiated expert algs Gradient 39 Alg Normalized WinnC Ada Boostquot 2839 Families of update algorithms Cont 39 Unnormalized Winnow Exp Grad Alg I Binary EXp Grad Alg any I 39 39 Bregman Hm 39 divergence 1 Members of different families exhibit different behavior 73000 1099 Wadmew cfHm 0190 B N grimeMary D i Wm W 39 29 LATER tress cmm 2 39 g M Pn io l o 197 i 557 m quot f I EkgwoaDx FmeH mew 7 H I mwwom We 1 Q BATCH Mewv 0F AerV stigvaw A MM AampMII CHOICE 04 DH E Q my c M c A L EXAMPLE 39 LMlggAra e g 11 2 PAQAMETExe f 52 g 53 v 29 601er Log g 911559 It 1 mm 3 aa39zAfz M6 T 439 21 W39Xg kit a gt 0 2 A Fl NIT D ion myavMpMTGAWvgz e OMLIUE Oimz PEI TieAL mmumwww mm v LWEM 4346 Jar V KKK egg Wei F EWWMEQQ RGG3 95 gawk f 209 91quot 39 u y 39 v39 I v 39 I V I 39 I 1 V 39 l I 39 h v 39 39 y 39 i w 027 6 39 i 9V W14 quot39 WT EE 43 EA is I a Law g amuss 51 me s 7 Law 1gquot fw 39 l 7 I 5 229 1 Wt 695 15 0N 391 gt gt relative loss bound for GD gt progruwquot22uw2etawxyxquot2l2 1 391 progrz5uw2 Euw2nwxyx2 gt p1factorprogr p12nx wxynxynxzwuw gt p22etaquot2wxyquot2xquot22etauxWquotXWXV p2 2n2wxy2x2 2n weW WCy gt factorprogrp2 O gt ewwxy euuxy gt p32etaA2ewquot2xquot22etaeuewew 2 p3 2n ewzxz 2n eu ew ew gt factorprogrsubsewwxyeuuxyp3 O gt to show that f ltO gt fp3aewquot2beuquot2 f21126w2x22neuewewaew2beu2 gt replace eu by worstcase eu gt di feu2feu cupped downward 2b gt soleusolvedifffeu0eu new saleu gt gfactorsubseusoleuf ew22n2x2bn22nbab 8 b gt diffgeta2geta cupped upward ew2 4 x2 b 2 b gt soletasolvediffgeta0eta b soleta 39 2 x2 b 1 gt hfactorsubsetasoletag 2czle9abew2 2fb1 gt solasolveh0a b sola N 2 x2 b 1 gt a andxb as functions of eta I gt solaeta 39 sola 1 gt solbSolvesoletaetab solb n e 1 2 n x2 gt bound Lusolblsola EDA2l2 l sola 39 Lu 1 ED2 bound 39 1 2 n x2 2 11 gt factordiffboundeta2 8Lux413 ED2 6ED2n x2512ED2n2x4 8 ED2113 x6 2 12nx23n3 gt Soletasolvedi boundeta0eta Soletaz l4ED2x24EDx Lu l4ED2x24EDx Lu 2 4Lux2 4ED2x4 9 2 4Lux24ED2x4 gt SOLETA EDI2xlsqrtLuEDx 1 ED SOLETA 2 x ED x Lu gt factorSoleta1SOLETA O gt bound2factorsubsetafactorSOLETAbound boundz EDX LuLuEDxJ IL gt finboundsqrtLuEDxquot2expandquot nbound w M0 at large ED2x22EijL MLu gt factorsubsetaSOLETAdiffboundeta2thus min at Soleta 3 39 8 x3 EDx Lu Lu EDxLu32 Lu32 ED 7 39 1 w A Wmm 39 aq39VE f r I 39 39 u a quot z r Hquot h k w 39 39 Wamwww I t r 7 1139 WM M X ms 39 Lu s i skC I gaLY l gt r z L A La L we fey I m Mum L n We c w TatMAW 5 gt8 MS S f 39 y 39 r a gt1 39 39 39 r 39 39 i I 39 39 II gt quot 5 l J 7 I I i I r quot K 3931 ENSITY HAWK we we t mgxrumw i M N E X E I wmwwm lt 3mm 1 7 L099 1 1 a 33 lt 39i I kg a rm ENMW e 4 39 3 39 wumm mm W W 1 39 1 MT em MM wum z I PMW 39 WW1 Aquot l bxWt 39 11 539 L4 WW 7 w 4 2 mg Inge43 vw 3 RAMi ampgt

### BOOM! Enjoy Your Free Notes!

We've added these Notes to your profile, click here to view them now.

### You're already Subscribed!

Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'

## Why people love StudySoup

#### "Knowing I can count on the Elite Notetaker in my class allows me to focus on what the professor is saying instead of just scribbling notes the whole time and falling behind."

#### "I made $350 in just two days after posting my first study guide."

#### "I was shooting for a perfect 4.0 GPA this semester. Having StudySoup as a study aid was critical to helping me achieve my goal...and I nailed it!"

#### "Their 'Elite Notetakers' are making over $1,200/month in sales by creating high quality content that helps their classmates in a time of need."

### Refund Policy

#### STUDYSOUP CANCELLATION POLICY

All subscriptions to StudySoup are paid in full at the time of subscribing. To change your credit card information or to cancel your subscription, go to "Edit Settings". All credit card information will be available there. If you should decide to cancel your subscription, it will continue to be valid until the next payment period, as all payments for the current period were made in advance. For special circumstances, please email support@studysoup.com

#### STUDYSOUP REFUND POLICY

StudySoup has more than 1 million course-specific study resources to help students study smarter. If you’re having trouble finding what you’re looking for, our customer support team can help you find what you need! Feel free to contact them here: support@studysoup.com

Recurring Subscriptions: If you have canceled your recurring subscription on the day of renewal and have not downloaded any documents, you may request a refund by submitting an email to support@studysoup.com

Satisfaction Guarantee: If you’re not satisfied with your subscription, you can contact us for further help. Contact must be made within 3 business days of your subscription purchase and your refund request will be subject for review.

Please Note: Refunds can never be provided more than 30 days after the initial purchase date regardless of your activity on the site.