New User Special Price Expires in

Let's log you in.

Sign in with Facebook


Don't have a StudySoup account? Create one here!


Create a StudySoup account

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

Sign up with Facebook


Create your account
By creating an account you agree to StudySoup's terms and conditions and privacy policy

Already have a StudySoup account? Login here

Topics in Computer Science and Engineering

by: Mozell Runolfsson

Topics in Computer Science and Engineering CSCE 4930

Marketplace > University of North Texas > ComputerScienence > CSCE 4930 > Topics in Computer Science and Engineering
Mozell Runolfsson
GPA 3.83

Ian Parberry

Almost Ready


These notes were just uploaded, and will be ready to view shortly.

Purchase these notes here, or revisit this page.

Either way, we'll remind you when they're ready :)

Preview These Notes for FREE

Get a free preview of these Notes, just enter your email below.

Unlock Preview
Unlock Preview

Preview these materials now for free

Why put in your email? Get access to more of this material and other relevant free materials for your school

View Preview

About this Document

Ian Parberry
Class Notes
25 ?




Popular in Course

Popular in ComputerScienence

This 31 page Class Notes was uploaded by Mozell Runolfsson on Sunday October 25, 2015. The Class Notes belongs to CSCE 4930 at University of North Texas taught by Ian Parberry in Fall. Since its upload, it has received 38 views. For similar materials see /class/229101/csce-4930-university-of-north-texas in ComputerScienence at University of North Texas.

Similar to CSCE 4930 at UNT

Popular in ComputerScienence


Reviews for Topics in Computer Science and Engineering


Report this Material


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: 10/25/15
my a ra fabi e WRy mn muaalmhwmg 11mm Mwnmvthyavmm mmmmam wmmwaddvfmwullmm anr39 svnlz n minute uh x hum WTme me Arti cial Intelligence in Games Stephen Gray Overview 0 Current Methods 0 Stateof thean and not so stateof the an 0 Current Research 0 Working hard to kill YOU better 39 0 open Problems 0 What s next 0 Brain vs Brain 1 Motivation 0 The motivation for researching how to make more humanlike Al in games boils down to two main ideas 0 Static targets are boring o Cheating bots are frustrating 0 Despite the tongueincheek comment in the overVIew the goal ofAl is NOT to kill the player 0 The goal IS to challenge the player and give a sense of accomplishment I Overview 0 Motivation 0 Why do we care 0 Turing Test 0 How much in uence should A M Turing have on the industry today 2 0 Classic Methods 0 Brain vs Brawn 1 L Overview 0 Review 0 What was that middle one again 0 Conclusion 0 What does it all meanl 0 Bibliography 0 Several people smarter than me is Motivation 0 Graphics are approaching photorealism 0 When all games look equally cool what will players use to separate good games from bad h Turing Test 0 Developed by Alan Mathison Turing in the 1930 s 0 Arguably was a central catalyst for the computer revolution 2 AlanTurlng lrl lElE Turing Test 0 Has been a long term goal and testing standard for Al for decades 0 Turing has also long been debated about its relevance to humanlevel intelligence goals o Is this really a good test of intelligence Classic Methods 0 Simple State Al 0 Rulebased o If tplayer within distance X a ac o lfhealthlt20 run away 7 v o Etc A 1 139 ginalDoomErain o Followed Brain vs Brawn model 0 Alve sim le so compensate by making enemies much more owerful andor numerous han the player 1 Turing Test 0 Definition A human judge engages in a natural language conversation with two other parties one a human and the other a machine if thejudge cannot reliably tell which is which then the machine is said to pass the test 6 Turing Test 0 Related problem The Chinese Room 0 Developed by John Seane in 1984 0 Computer in one room person in another 0 Both asked to translate a Chinese sentence 0 True Al would be indistinguishable from the person Current Methods 0A a Gold standard for path a I n nding algorithms mmquot quotmquot m 0 One ofthe few algorithms f developed in academia that was embraced by the mainstream game community 0 DA 0 Extension to A39 designed to limit memory usage 0 Performance poorer guy 3439 in action 5 because algorithm forgets past choices and can repeat sub paths Current Methods 0 SMA Simplified Memory Bounded A 1 0 Improved version of IDA 0 Only forgets once memory allotment is exhausted o Tries to forget in an intelligent way worst looking path choices rst L Current Methods 0 Neural Networks 0 Anothernature based algorithm 0 Works by evaluating weighted feedback levels similar to the way neurons work in the human brain 0 Requires human input to tell the system what was a good choice and bad choice 0 Implemented in the game 0 Not embraced mainstream Current Research 0 Motivated Reinforcement 0 Complex enough to have its own lecture Cover basics here One size ts all System learns on its own with motivation function Implemented a test in SecondLife 0 However perhaps the most intriguing advances in a life as it relates to games is interagent communication agents working together to achieve a common goal 1 o Thisfalls in line with my own goals for this class I Current Methods 0 Genetic Algorithms 0 Works by breaking problem down into genes 0 Usually binary genes 0 Each cycle of the algorithm breeds mutates and culls the population 0 Culling evaluated via a fitness function 0 Solution found after certain pre de ned fitness eached o 39Requires large amounts of resources and time o Requires hand tuning to get correct tness function 0 Implemented in Crea ures t 0 Example htlltsfelkcvutczlxobitkolal Current Methods 0 Reflexive Agents 4 0 State Machines and Rule Based Response 0 Fast but brittle 0 Must be custom built to each application 0 Learning Agents 4 0 Genetic Algorithms and Neural nets as discussed previously 0 Slow but flexible 0 Not embraced mainstream due to limited CPU cycles l for A Current Research 0 Fuzzy Logic 1 o Iouderthanabomb Iouderthanabombcom designed a fuzzy logic editor speci cally for game design 0 Dynamic Difficulty Adjustment 3 0 Dynamic change of dif culty via economy manipulation life increase enemy decrease o yirtually unperceivable to player but familiar players reported a greater enjoyment 0 Change blindness prevents most players from noticing 6 Basically that a player will not notice even semi major changes if they occur during screen wipes or other transitions i Open Problems 0 Dynamic content adjustment 3 0 Game provides subtle help if you are getting hammered 0 Adjustable assistance rate depending on desired dif culty 0 Designed to improve player experience by gradually becoming less helpful as player skill increases L Open Problems 0 Artificial life 1 cont 0 Collaboration Outnumbered Mobs will attempt to ee and regroup with allies o Teamwork Mobs effectively work together Strong tank like creatures up front block passage to weaker ranged in back L Review 0 Current Research 0 Motivated Reinforcement 0 Open Problems 0 Bridge the gap between academia and game companies 0 Re ne algorithms to be acceptable for realtime use I Open Problems 0 Artificial life 1 o Emotion Some of your troops in RTS will retreat spontaneously 0 Fear FPS mobs may flee in terror after you blow their leader to shreds o Tactics Mobs that learn your tactics and compensate like a player Review 0 Turing Test 0 Who will lead us in the next 50 years of research 0 Classic Methods 0 Brute force to compensate for stupidity 0 current Methods 0 Aquot 0 Advanced learning not yet widely accepted 0 Simple aggression algorithms Conclusions 0 We have only scratched the surface of what is possible 0 Game developers are realizing that engaging enemy behavior is becoming key to player interest x 0 Increased CPU time is being allotted forAl calculations up from 5 in 1997 to 10 in 1 Conclusions 0 Much work remains to be done to make efficient algorithms 0 Schism between academia and mainstream game companies must be bridged before true advances can be made Bibliography Kathryn Merrick Mary Lou Maher Motivated Reinforcement Learning for NonPlayer characters in Persistent Computer Game Worlds ACM International Conference Proceeding Series Proceedings of the 2006 ACM SIGCHI n3 2006 Alexander Nareyek AI in Computer Games Queue Archive Vol 1 Issue 10 p5865 February 2004 ht lenwiki ediaor lwikiITunn test httenwikiediaorlwikiImaeNeura networksv Bibliography 1 B C Bridger Chris S Groskopf AI and complexity ACM Southeast Regional Conference archive Proceedings of the 38th annual on Southeast regional conference p5155 2000 Varol Akman Patiick Blackburn Editorial Alan Turing and Arti cial Intelligence Journal of Logic Language and Information Vol 9 Num4 p391395 October 2000 Robin Hunicke The Case for Dynamic Difficulty Adjustment in Games ACM International Conference Proceeding Series Proceedings of the 2005 ACM SIGCHI p429433 2005 g I I R e39sults 0 Future Work an doseto 000 no of 511 210 04333Q 98 operatxo s T S39ar ou tisappro matew 15 98Aquot fl V 81quot00QO Impossible solution set An NVIDIAJBQO GTX 512 is capab 200 GFLOPS and the current 1106 DIA 8800 GTX is capable of sustaining 330 GFLOPS V o It s GeForce SB GPO is capable of billions of These Shader languages are mostly u Sedin calculations per second can process a minimum graphic oriented GPU Quara 39ons 39 of10 million polygons per second and as over 22 million transistors compared tothe Smillion found on the Pentium III mnmdmr uwnumm t Amunm mm m Immummmm A mu H Um mam Here s m x mm A ank mm m mama qu mmpmng a Program mamas ymgmmbvamlingumamnmmlh aucnnuuwnn mama mm mum may x 1mm Ilve c Mm cum mnrwlmuasumq nnnnumx mm nu F IE on unuw a mm m m 3 mm Al39l mmyalu c m I magnum mum 4194 H iskowwimlude the openGL c5 nummmrn mung mm m lmw m m ude m nmun 9 Cg mmam Aw mum cgmm mome duewimmlanrmu gymnanwmmom mmauwmu mo 0 39mism m win 51 q mnrw ulfvlnynxuvur mm at m 39nymmmn meaning a Progra m 54mm mam msmn n 0mm m me 15 Aawm rm l mm mu mam quotmmWhimmwzu wwwmm mWWJWMMW whammy m cmrund m wralnmrvrduv mm mm m 115 Mum39s mummeeux mud u uxmmmw mqu a u 39textures gist rs in PU Wnte vaxuegto Texturequot R ad va ueg back39tq CPU mtgwwmahmnamum mm as neddeKsggggunumnavzhum c nsava Ca Bqunc i SEE samp g nude NV mA Fnr EBpU u ZSEVENGVapmcstmssngUM 755m Mem njy Menacewcunm Mammy I MemnWBanmmh r 35 ms peygemnu Fm Rare 5 3 Emmy TeXe S nay Semnd Venex Pvncessmg 7 Gun MHHnn Vemues pev 5mm Supevsca av 1 Eerpehne Avcmedure nmpmmmmw wmummmm mum mmWmmummu r quotmum WEE w niumv as d mulwcnr apps Wh39ele nwywwweaas Hhe same lune 39 e n 5mm MW Ruemam augments m was an en NV D AGPUSYeamvmg UDMechnu ugypvucess huusandau bmsnammmm hvsadsswmuhaneuus enabhng a hwghev capamty m Muvma pn r nw mm mum be mum N Wm mm a La t D scuismg dquUDM hnb u m Geneer Pur nsa39c mpummn nn Gr GPGPU V We Huusmn 51mm Unws M Na hcss dewmh 4m umg plum Pmcvssnrs amass span 5 devbmgvammmg 3mg pavs b pmaassngjmuy w spawn Pe nrmz K mm uuslun ubh am 5m WI p d nus nceTuhInu Amuus ed examI2 Dumer Gaddeke Unwmw39mrsamm Gevmany Dwspussesth basmmncep su m uves a wmmmmmn m DesHhestemmquesthm mm mm epu based Wp ememm m o Discusses basic mafh opt rafi onzusing GPGPU l39 a 39 39 u 92 new we hm llmmdurangwbillcumlN Queenantml tt Ilwww ad 3 v k OOIec mslalLec u 22 e BOOKS GPU GemS39ZUOE GPU GemE 2 2005 RealTime Water Rendering CSCE 5933 TDpiES in Game Deveiuprnent Jeff McGee AprS 2mm Simulation of Water in Games Mutivatiun Endiesntwatev ave mmmnn inaiitvpesntSD game settings Reaiis c vepvesematinn is chaiienging but important mv piavev immevsinn Tvvu Aspects uf Water Simuiatiun De nmgthe uvtace m We unwemnmmnan physmi u M m mmmnm quotmi Rendeving the 5mm mm m W5 ms mmth Gama wmersur ce Rendering the Water Surface for Above Water POV 4 quotI 2 l i Sum mm Components mm Wm mmvmmnmnama m lt3 sumatwaca vanems mindm l 41 Computing Each Component unmnunmaemmwnm wimstnemmn mm mmmmunuwm m mm mnma iivM camnv mm mm 7 mnmnemmmncnmmmnvmmannngmnmwntnemn mm and mumvivm vmne nai wmvanem usmsssepmwmmnanannsmssan inameri 3 2 sun mm nnsmwnum Re ection Component g Computing the Incident Vector for Re ection Wheve isthe ve ecled iighl coming an7 5 st i thnwmw i mmmm WNW a WWWth mm at e cm cm Weii Knnvmiawntvetieclinn a z a Veda innn x Computing the Re ection Ef ciency Hnwmuch ntthe modem iight is Ye eder x n mnean gt7 nnsmmn emuenCV Fvemei Equation were were sm2 away ta n z Concerns Amnbmneeqmn new mm nnsmmn an wncn we mivM m mnemiseneed uniesswevianm indude mm mamanmisim Options m cawme iaakuvubie mum hrme wmerair mm and index mnth mm an r wpmxmewasianunmnmmnnaepenasan m ncidemanvieovmecammn see nenii Approximations to the Fresnel Equation remit mm maximum 3 mm m m t in WW m m 1 g mg Fresnel Ewen mm mm wmersur cehead anmmmum isvnmnlvnnsmd uvm summm my Mm is vnmnlvm eued Appmmmns mam base Mel masmsmlumm mm l Mignon Wm FarEWISaugn Computing the Incident Light Color for Re ection What gm is coming m this medium Wenavemund A mWmmwmmmmmmm x mmthmmwmlMWMM n L a MWWMWM W 1 z 1 5 5 r m L a nallv m mm m L z Wna acansdev mmlmmabm Mum EubicEmmnmemh v Mum Madned minimumquot lnudemllvmmmsaums Mum 3mm SpecularMadelmimumn mm sources Mum mm Sample mm mm mm sources Re ecting Objects via Cube Mapping Ve Appimmma Water mm remnnuemwmmm mammall Wienismallmlmimmme ammammmms Dnlvsumundmvsdeviaedinmeswbaxam 54mm mm m m badvatwmer Wmquot Britusecubemivvmvhrm eamvtm mm was mm mountains at Mm mm mm anapassmymwmc notetum mammamepm Sammie Example of Cube Map Re ection SWBuxRe emaanwacnkauSl Wmmmmmmnm Nalacalanmmms mg etCharefen Re ecting Objects via Planar Re ection mm W Planar re ection basics 7 mm Givenaplane mumquot them euedsmnecamwandsmmesmnemma a Mm amenabamthEDlane and mmummmmmmm r a Rasmruheanvmal em amewmmm wmermsh r a usumwwm mum wmr vmemm emplmmmmm mm mm Wammmm wmrmmmmwmwm W 5 mmEDlmenaml Accounting for Bumpy Surfaces in Planar Re ection mm W Yex uvesamplemmlacemem 7 mm Wecansull use maplanarmledanmmm islanvaswecawmean mm quotmesmequot We mmsvlmmmsusmmmm and mum ume vasman mm mm m svumeableuvasman mmmmme m mam displammem Wm 52mm Computing Reflection Texture Displacement Texture Displacement see appendix gear 25 sauce cibieci tanquot 7 1Qc052 dxsp V m Cornrnori Approximation dISp 0 sin a Approximation inaccuracies Example of Reflection Texture Sampling Usirig Previous Displacement Approximation dixpc sine lquot c mm c nus Modified Plariar Reflection SOUSaO5 e ensiani cciniiciis wiggie inine ieiieciee Dialects cine Size musiiii aiii e Tweak ccinsiani ici inikguud ibibciin bux neai and mciuniain iai Summary 7 Beneiii can be used in ieiieci bciin neai and iai cibiecisWiin decent accuiacy e Limiiaiicin Requiresrenderinglheiei39lediunlexlureeachirame Musiayciieiaige eispiacemeniswnicn iaii bui cii ayaiiabie Iexluyel Reflecting Sources via Standard Specular Model For DirectioncDefined Sources 3 3 surface nciimai gnen areflediuriViewveclui given rdirecliunal iigni sciuice given gtgtxgt h chalivedur bisedinglightveil 5h chalHurnurmalangle norma1128l r E amaleiial specular expunenl LWM Mme M amalerialspecularmulliplier Specularlviodel basics e Cumpules iniensiiy ciia specularly ieiieciee iigni sciuice wncise eiieciicin is eeiinee by a yeciciiriciminewaieiiiagmeni Specular expunem eeieimines newcicusee isine ieiieciee iigni nignibiwaien i e 1E Specular muiipiiei eeieimines iniensiy ciiine peak ieiieciicin ieiaiiye ici incieeni iniensiiy neaipeneciiciiwaienie M i Example of DirectionDefined Source Reflection image from SAGE Paibeiiyosi 7 Sun ieiieciicins generated by sianeaie specular mudel Reflecting Sources via Texture Sample Boosting Fo TexturecDefiried Sources Reilediun Texture TexturecDefiried Source 7 Same ciiine incieeni ling sampled mm mm ieiieciicin texture cube cii planar migni be turning mm a bright sciuice like ine n su e Sucn iigni is puurly iepieseniee in ine scene Wiin ciui iimiiee dynamic range V liEIW else Euuld We slave raighl allhe 5qu e We can ci leas1 make ine ieiieciicins brighter and mciie ieaiisiic by bcicisiing ine incieeni ling L 1 xafumfeR r TB Example Boost Method KryacnkoOS e s delected by ccimpanng same 7 measuie ciiiis iniensiiy neie ine iee mmmeo a me D 3quot cumpunem in a iniesnciie yaiue Euus1 value ensureslhal reflected light iemains bright aiiei ieiieciicin eiiiciency is applied R rincideril iee iniensiiy T quothiesh ld iee iniensiiy m 97 B abuuslmullipliera Example of TextureDefined Source Reflection Original Texture Sample Boosting Kryachk005 e Engni sun ieiieciicins by bcicisiing texture sample Computing the Incident Vector for Transmission Whev quotBrim e isthetvansmit ed iieni caning e 3 Transmission Component e Sneii siawmvvewa ian K sm Veemiieni if lt i The WaterAir Refraction Effect Computing the Transmission Ef ciency nannuen atthe incident IigM istvansmit ed7 Use Fresnei again n ereeuanen gt7 nnsnieean eneenei riesnei Eguatmn Light mnsevvatian n e Shun2m e i n eawnneiineeneeeneneeneieiweiei newneniieinesene ineinemiieenv imbue n beiaw appendix e We unaware inanenneiieianeneiennwim enanei Mei wmvmmn Computing the Incident Light Color for Transmission Incident Light from the Water Volume What iieni is mming nan inieeiieeiann CampIeXPmcess Wehmmm A me e ineineeeniinnieineeunaiewn eeinn e nemennnwnenieneiiieeiiinanineeei r 23933quotquot jggggg mmsm i neeimiineneeen innineiwiinneiieneenen gt7 s neeepei ieeeeninenneiinenenene in L neee eepene nine Wei mam Wemmm i b needinveieiimnimi semieneie neemeinmneneieenin L e 1mg 5min W e Mineiiiinemnmeeiim LnL APPVDX Singiecaim e peenbeeeniieeinneiiweewmpneie Whammm39 L eipeneuieiweeimi e Inudemiivmm eeeiienni n wiune Temnique sinvie Coiar nique Aew Anpmx AeWDEPendem c Yew nepeneen Coiar e e ineeeniimnabieee iemnique FWVmV iemnique inneiieeianiemsenenv WWW wmw mwm biteineniaakinveieeuiinmiuneene bnvmrvmeniaakmvammwaws m e Dee e g lncident Light from Objects Hawistneimensityveducem Again ACamPlemecess smenneeneebsmieneemismneniee iimmn em me eneeee livM eeenmne We pnnmneeimenumn Appmx nagging new been nememneaniemm bed in e an bedmcunm eeennne unieem men is eeser and epvmpneie inen men in wilerdisunce is sneii and meme miew ween causes Facusnv anode hcusnv atlivM ravsduem mm mm slrbce Meas llumnmianat undemiembiess sumnv Dual Men 7 wpmnmiensm wveredhere TwicallV an em mmi dimmer lncident Light from Objects oeieniining same ahie Win atransmissian texture W TranmissmnTex ureSam lin 5 V e m m e Wmvnm eisieeenemeeeee me a mmv meienenee Steps URenderal below weereneesimnemnsnseimem e maeemneenmneweernes e msenpieeneeemusinwasnenpiusmanneispiemem ne vmblema ndinvme wmadisvlaeememissmilarmmmatme refunanmxmre Neesieeeismsses WhatabomUndevv levCubeMappinW Net museui sneeeismuneemiemnieesm amallvcawlemlyabsmred Computing Transmission Texture Displacement TenuveDi lacementseea endix smta en tan 7V0 17cas disp e in Acamman Anpwneiian 7 mm W nmbleemumat memene Wivle absemd NEITmeavenllsmlmv vlervtma De ne ieiem like n en eisieeenem ivvmximian inen enviimnemnsnssimem Wee smiquot Example of Transmission Texture Sampling a ee sinsquot Using PVEviaus Displaoeniem Anpwneiian e am e an e um mermnsdisabled mienesseiebiea illevens nensnissian Texture Sampling SDusaDS e we ammll 3mm enesiseiemnissnv we uivvle eeeieeisewnenveneuvnernessmiens Sunnew 7 men an tense mam neerene muesli demmacwmcv 7 mm neeuieseneenm mennsmssanmmm em name Muiawidliwe eisieeenemsinemiimiminieemi Example of a Final Rendering Sousa05 er s 39 Summary of Water Rendering Tine 2quot Half uf Water Simulatiun Omianisadetinedsuvlsoemesnatvenexpastmnsandnamials came he a vial grid at semiiing hump vectors m saneining mum more complex Renderingtne Surface Pevoewed suits2 calm istne sum eneieeiee ieni and iiensimee iieni anscallamandgaadap Xi atian allawustadeterminehawmuch ieieeiianmensiissmn We have and n Wnal divediansta look in sum22s Finding tne incident Lignt Mas eneienene sap ism eeieniine ine campasitian at lignttnal is ineieeni upon ine water suns2 Techniquesinvulve sometami aisnpine quotuni environmental te uves with adlusmems such es msne bright swings m lugging uneemeien Appenmx Suurces JensenmlLasse Sta Jensen Ruben Guhas eep39Wa evAmmauunand 39endennu hugHWngamasmvacurngdceZEImensenensen m mm mm hsmmu llJuhn smum A exWachus mesEvennan RendennuOceanWalev DwectSD shaeevx VenexandPixe ShadevasandTnckS WDVWavePubhshmg l z MachusnnmexwmusJuhnxsmme Chnsoat R39wmhng Re ecweand Re acwevvatev DwectSD shaeevx Venex and New Shade ms and mks Wumvvave Pubhsmng znnz Tesseneunm Jew Tesseneum Smwalmg OceanWa ev Smwawvg Name S GGRAPH Cuuvse Nmes 2mm Kwachku l Vun meme Usme Venex szmve D smacemem 1m Reahsm Wa ev R2nd2nn GPUGemsZ PmuvammnuTEchnmuesfuvHmhrPe uvmanceGvapmcsand Geneva Pumuse Cumpmatmn Admsunrvvesbv m5 sensanamaeuSeusa oenencRenamequots mmauunwpu Gemsz Pmevamme Techmquesfuv ngnPe uvmance Gvapmcs and Geneva Puvpuse Cumpmatmn Admsunr Wes EV ZUUS avsun JereVW Nurm JueSchembem JasmnCu E SAGE A My May 252 um edwsagek zane PameWUB Wan PameW Enk Swmp e Academe came Engm Appenmx Re ectmn Texture Samp e Dwsp acement Equatmn 8 4 525 725 n m g m 251osa750525psma 95m 25 and 7V 150525 51W Appenmx Transmwssmn Texture Samp e Dwsp acement Equatmn abc180 5180 79mg sma smb suuvce meet 0 sma 75mg V m5i cosaicos i sma 0 414575 Lana V l cos575 dxsp Appenmx Pruuf uf Equa Re ectmn Emmeney Wm Shared Vwevv Vectur z Re emun Transmwssmn GWEN 9W 92m 2quot f 49 75 at 5 a msxn msm 2m nasm 1 9 stn me we 49 7 3 3 mzewe 5 dl lt n dl Neural Networks and Particle Swarms By Jeremy Nunn Overview 39 Neural Networks NN 39 Particle Swarm Optimizations PSO 39 Training an NN using a PSO Neural Networks 39 Arti cial Neural Network ANN orjust NN based on the human brain 39 Connected nodes like a graph 39 Takes some inputs gives outputs N n quot NN Background 39 Branch of artificial intelligence or arti cial life 39 Dates backto the 1940 s NoCulloch and Pitts developed the rst neural model 1 o 1962 F Rosenblatt develops the perceptron model which is useful for solving very simple pattern classi cation problems 39 Interest disappears in the 70 s 39 Since the 80 s has seen renewed interest because of better network designs better training algorithms and faster hardware Some NN Vocabulary 39 Neuron a node in the network 39 Activation value the computed value ofa node based on weighted inputs 39 Threshold the value that determines whether a neuron will re pass its output onward 39 Training the process of tuning the network by adjusting the weights the idea is to get better results NN Examples It NN Math n Uand rtsthreshuld TU We s Luukrng at some neuru calculate the activatan value AU a AU WM qux2 WM 23 Where W is the Werghtvalue er the edge frurn rnputr m neurunr The neurun fires uutputrsl u rfAU gt TU usually other amrvatrun functions EXl be AU ur ma st such asrust alluvvrng the uutputtu pprng Rm r e sigmoid Why use NNs Have been used in speech synthesis diagnostic problems computer vision Can be used when a rulebased expert system would be too complicated to design Can be used to recognize complicated patterns even with noisy input Can use parallel algorithms Why use NNs con t Ford Motor Co used NNs to diagnose engine mal Jnctions Boeing Aircralt Co use NNs to nd a best match for manufacturing a new part given existing parts being manufactured 2 More stuff sales forecasting customer research risk management which includes credit scoring and underwriting Particle Swarm Optimization Screen Sm m the Serbs simulation round at is In the catego PSO Background Developed in 1995 by Eberhart and ennedy Derived from models that simulate ocking and schooling see boids and oys Similar to Genetic Algorithms and Ant Colony Optimizations o ALifethatuses biological techniques to help solve problems PSO Overview Each particle represents a solution As the particles y around the search space they are evaluated given a tness They remember their best and they are told their neighbors best These two locations will in uence their movement PSO Algorithm Velocity as a e 9 c pbext e pm2mcz72 game prawn i where c1 and c2 are learning factors and r1 and r2 are vectors of random values between 01 Position present presenter 2 P80 Algorithm For each mice lnnialize mice are Du For each mice mime Muss value iiintmntss wine is be eithan int bstinness valueOBest in new cuiientvalue 5 in new pest End Change int particle wnh int bsiirintss rut Di allthe mieis 5 the vets For each mice mime mice may accaidinv mm m we vamcle mm minim emirantz End While WWW neiatians m Minimum trim enen e m mm m PSO Tuning Tnere are a limited number er parameters thatneed tn be tun Number er partieies Maximum velucity Learning raeters step Bundltl ri inertia weignt Determined bythe prublern Particle dimEhSiDri ange PSO Application Any function that you need to optimize 7 Even Witn discuritiriuuus surfaces 4 Can be used in multiobjective multiconstraint cases Evotving NNs Functions where the optima move overtime Many realworld problems 7 Reactive puvver and vultage eentrei e lngredient mix eptirnizatie 7 Pressure vessel design Training a NN using a PSO Training a NN using PSO A NN s effectiveness is in its weights assuming the topology is correct Other methods exist to train NNs butthey can be slow andor di cult to implement GA s backpropagation etc How can we use a PSO to do this Training con t Every edge or weight in the NN is an element in our particle if our NN has k weights our particles will have k dimensions The fitness value is how well the NN performs on the test data We run the P80 fora given numberof iterations or until a minimum error is reached My Project Briefly Using a NN to classify songs into genres Training the NN on 500 song samples in wav format lnitial setup will use 5 outputs genres Once the NN gets all the test data right try nontest data Compare the NN results with human results to verify correctness NN PSO Uses in Games NNs can be used in just about any game Although other options may be better Card games lllty hand opponent s bet time he took to act e c Board games Go Chess TicTacToe AgentAl FPS RTS References Neural Networks httpIuhavaxhartfordedulcompscineuralnetworks tutorialhtml Russell Ingrid Neural Networks httpMMANdocicacukndsurprise 96ournallvol4cs 11reporthtml Stergiou C Siganos D Particle Swarm Optimization httpNwwvswarmintelligenceorg Particle Swarm Optimization Kennedy J Eberhart R romm rocl E lnt l Conf on NeuralNetworks IE E Service Center Piscataway NJ lV19421948 N Fquot F References 5 Boids httpwwwred3dcomcwrboids Reynolds Craig What is Procedural Generation Procedural Generation or Procedural P roce d u ral Ge n e ratl o n Synthesis describes algorithms and methods to create content on the y as opposed to creating it before distribution Joshua Taylor 599 quotis 9 Procedural generation 39 039 use Things You Can Procedurally Generate Textures Textures Environmental Effects Procedural Textures give more diversity with less memory Can be used to avoid obvious tiling quotModels 5 J5 W thesis P39erlin noise Animation 3D Textures Environmental Effects Textures can be 3D as well as just 2D Things like clouds exterior lighting wind Surfaces take a slice through a 3D texture grassy rev 5m kev 9 0 Can be created Curved suf ces takea curved slice procedurally to enhance the environment le aving fewer se ams Most of thisquot s e 39alusesofeither 3 D te 7 e V j m r proc39edural textu39res shaders art sl39ce 39 V 39 39 l quot ngines Clouds and Sky Models Mesh subdivision and mesh reduction hange the number of polys Meshes can be created around a data set e m rching cub salgorithfn Animation Ragdoll Physics might be considered procedural animation Particle systems might also be considered proceduralanimations 39 Q 39en Interact with th e39 physics s y m Sound Computer generated voice could be used to give every character a voice without a voice actor Animation computer decIde how to m walk ghtdance rt39a 39rnatio n clip 39 can b ebl39ende quotto oother Given a mesh with a skeleton let the ake it Behavior Al Al should be procedural by de nition Some Als are just hard coded nIles Some Als can learn and ada t or ek rple chara39cter m39ood dealt39w h procedurally Dialog Instead of having hard coded lines of dialog you could code what to say and procedurally generate how to say it Dialog Would be different from play to play andcould adapt tolth eisituation39 Levels Most common thing to procedurally generate along with textures Done often but not oftendone well Indoor Settings OfficeResidential Building Other Manmade Spaces Natural Spaces 0 Exotic LOcatio39nsgt Story Have the computer make up and tell a story Might be write a text file or set upa game scenario 0 ProceduralGameDu39ngeOn39lllaster Types of Levels Galaxies Solar Systems Plan39ets 39Planetary Terrain Clutter An indoor space in real life is rarely if ever an empty box A living room will usually have a couch a TV and manyother random p39 39 of ell as other rand My Research My Procedural Clutter Improve procedural level generation by adding Add attachment points to the room two maybe threellhiHQSi Add attachment points to objects Procedural Clutter so thata room can look like pIace objects at attachment points a liyingLroomwithoutylookingulikejhe gameliying rsively39usin robabilist N My Procedural Clutter in 2D gyagumn Room Relationships Built on the can be adjacent relation for room templates For example the kitchen can be adjacent to the dining room ela 39 Irregular Rooms By dividing space into a series of boxes and assigning each box as either solid or empty you can describe many interesting room shapes By applying models to the intersections of the grid y an round the edges andma39ke the Conclusion Procedural generation is any method or algorithm that allows the computer to create content for you an be created some Room Relationships Applying templates to rooms in a way that the template relations hold is an NP complete problem a superset of graph coloring Irregular Rooms The other common approach is to use prefabs to build levels This allows moredetailv but is also more granular hislcausesxlevelszto39looklAveryzsimilar39 Withoutusing any pre a s Augegmi 39 References Automatic Furniture Population of Large Architectural Models Kari Anne Hoier Kjolaas Beyond the Horizon Stefan Greuter Nigel


Buy Material

Are you sure you want to buy this material for

25 Karma

Buy Material

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

Steve Martinelli UC Los Angeles

"There's no way I would have passed my Organic Chemistry class this semester without the notes and study guides I got from StudySoup."

Jennifer McGill UCSF Med School

"Selling my MCAT study guides and notes has been a great source of side revenue while I'm in school. Some months I'm making over $500! Plus, it makes me happy knowing that I'm helping future med students with their MCAT."

Bentley McCaw University of Florida

"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."

Become an Elite Notetaker and start selling your notes online!

Refund 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


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:

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

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.