꿈꾸는 Agent

ԷԲח"HFOU 8POTFPL+VOH

  8POTFPL+VOH $JUZ6OJWFSTJUZPG/FX:PSL#BSVDI$PMMFHF %BUB4DJFODF.BKPS $POOFYJPO"*'PVOEFS %FFQ-FBSOJOH$PMMFHF3FJOGPSDFNFOU-FBSOJOH3FTFBSDIFS .PEVMBCT$53--FBEFS 4DIPPMPG"*0QFSBUPS  3FJOGPSDFNFOU-FBSOJOH
0CKFDU%FUFDUJPO $IBUCPU (JUIVC IUUQTHJUIVCDPNXPOTFPLKVOH 'BDFCPPL IUUQTXXXGBDFCPPLDPNXTKVOH  #MPH IUUQTXPOTFPLKVOHHJUIVCJP  :PVUVCF  IUUQTXXXZPVUVCFDPNDIBOOFM6$N5Y8,EIM8W+6GS3X 

"XBSET Ӗ۽ߥ҃ઁޙചߊ੹؀࢚ "*ӝࣿഄनࠗޙ؀࢚࢚ࣻ Ҵഥ੄ਗ੢಴ହ੢ࣻৈ   .PEVMBCT'JSTUDMBTTJGJDBUJPODPNQFUJUJPOT SE  .PEVMBCT'JSTU3-$PNQFUJUJPOT  TU 
  .PEVMBCT4FDPOE3-$PNQFUJUJPOT TU 0QFO"*3FJOGPSDFNFOUMFBSOJOH3FUSP 5PQ

  1SFTFOUBUJPOBOE-FDUVSFT 5FOTPSGMPX,33FJOGPSDFNFOU-FBSOJOH ऱ੉Ӓېݠ%*:ъച೟ण ա݅੄गಌܻ݃য়ڣٯڣٯٜ݅ӝ IUUQTXXXZPVUVCFDPNXBUDI W#.*JSCW2X
ী੉੹౟ܳࠗఌ೧5IFSJTFPG3FJOGPSDFNFOU-FBSOJOH IUUQTXXXZPVUVCFDPNXBUDI WVI9V"L3G.2 TUೣԋೞח٩۞׬ஶಌ۠झ3VOOJOH"* IUUQTXXXTMJEFTIBSFOFUXPOTFPLKVOHEFFQMFBSOJOHDPOG ౵੉௑௏ܻইੋҕ૑מगಌܻ݃য়੄Ѣ੄ݽٚѪ IUUQTTQFBLFSEFDLDPNXPOTFPLKVOHBEPUJTVQFSNBSJP XJUISFJOGPSDFNFOUMFBSOJOH ೠҴ੿ࠁ୊ܻ೟ഥъച೟ण੉ۿҗपઁ IUUQTXXXTMJEFTIBSFOFUXPOTFPLKVOHSM ా҅աۄъച೟ण੉ۿҗपઁ IUUQTXXXTMJEFTIBSFOFUXPOTFPLKVOHSM

https://medium.com/@madanram/you-only-look-once-aaf841df2c7b

http://www.relationclientmag.fr/Thematique/techno-ux-1256/Breves/Chatbot-B2B-qualite-service-peut-aussi-passer-machines-318813.htm#myUmcA7SSRYBzhzk.97

WHAT I’VE DONE REINFORCEMENT LEARNING

R.L REINFORCEMENT LEARNING Reinforcement Learning

(PBM 8BMM "HFOU "DUJPO "DUJPO "DUJPO REINFORCEMENT LEARNING CURRICULUM LEARNING

IMITATION LEARNING *NJUBUJPO-FBSOJOH 5FBDIFS 4UVEFOU REINFORCEMENT LEARNING

R.L REINFORCEMENT LEARNING https://ubisafe.org/explore/demeanure-clipart-prior-knowledge/ Meta Reinforcement Learning Exploration Reinforcement Learning

HOW ANIMALS LEARN

ALL ANIMALS HAVE THE ABILITY TO LEARN - ݽٚزޛ਷೟णמ۱੉੓׮ -
ৈѐ੄न҃ࣁನ݅ਸыҊ੓ח৘ࢂԘ݃ࢶ୽ژೠ೟णמ۱੉੓׮ - ݠ୍ܻࣻ߈ࢎIFBEXJUIESBXTSFGMFYਤ೷ೠޛ୓о੓ਸѪ੉ۄ౸ױীٮܲ߈ࢎ೯ز - ৘ࢂԘ݃ࢶ୽੄ݠܻܳѤܻ٘ݶੌ੿Ѣٍܻܳ۽р׮ HOW ANIMALS LEARN

HABITUATION HOW ANIMALS LEARN 'JSTUUSZ 4FDPOEUSZ 5IJSEUSZ

LAW OF EFFECT - &EXBSE5IPSOEJLF - -BXPGFGGFDUযڃ೯ز੄Ѿҗо݅઒झ۞਋ݶ׮਺ীبӒ೯زਸ߈ࠂೠ׮ ߈؀۽݅઒ೞ૑ঋਵݶӒ೯زਸೞ૑ঋח׮
- 3FJOGPSDFNFOU ъച ੉੹ীੌযդ೯زਸ߈ࠂೞѱ݅٘ח੗ӓ - 1VOJTINFOU ୊ߥ ੉੹ীੌযդ೯زਸೖೞѱ݅٘ח੗ӓ HOW ANIMALS LEARN

EXAMPLE OF THE LAW OF EFFECT HOW ANIMALS LEARN

HOW HUMANS LEARN

INTERACTION WITH ENVIRONMENT &OWJSPONFOU &YQFSJFODF -FBSO *OUFSBDUJPO HOW HUMANS LEARN

HOW HUMANS LEARN? - 3FJOGPSDFNFOU੉੹ীੌযդ೯زਸ߈ࠂೞѱ݅٘ח੗ӓ - 1VOJTINFOU੉੹ীੌযդ೯زਸೖೞѱ݅٘ח੗ӓ HOW HUMANS LEARN

REINFORCEMENT LEARNING

REINFORCEMENT LEARNING &OWJSPONFOU &YQFSJFODF -FBSO REINFORCEMENT LEARNING *OUFSBDUJPO

LEARNING 3FJOGPSDFNFOUMFBSOJOH਷3FXBSE ࠁ࢚ ਸ୭؀ചೞחBDUJPO ೯ز ਸࢶఖೠ׮ -FBSOFS ߓ਋ח੗ חৈ۞BDUJPOਸ೧ࠁݴ SFXBSEܳо੢֫ѱ߉חBDUJPOਸ଺ח׮
ࢶఖػBDUJPO੉׼੢੄SFXBSEࡺ݅ইצ ׮਺੄࢚ടژח׮਺ੌযաѱؼ SFXBSEীب৔ೱਸՙசࣻب੓׮ "DUJPO ׼੢੄ ࢚ട߸ച ޷ې੄࢚ട 3FXBSE ޷ې੄3FXBSE REINFORCEMENT LEARNING

MARKOV DECISION PROCESS "DUJPO "HFOU &OWJSPONFOU 3FXBSE A t R
t 4UBUF S t R t+1 S t+1 REINFORCEMENT LEARNING

AGENT "DUJPO "HFOU &OWJSPONFOU 3FXBSE A t R t 4UBUF
S t R t+1 S t+1 REINFORCEMENT LEARNING

ACTION "DUJPO "HFOU &OWJSPONFOU 3FXBSE A t R t 4UBUF
S t R t+1 S t+1 REINFORCEMENT LEARNING

OBSERVATION, REWARD "DUJPO "HFOU &OWJSPONFOU 3FXBSE A t R t
4UBUF S t R t+1 S t+1 REINFORCEMENT LEARNING

MARKOV DECISION PROCESS &OWJSPONFOU 3FXBSE A t R t S
t R t+1 S t+1 REINFORCEMENT LEARNING &TDBQFGSPN UIFCPY 1PTJUJWF3FXBSE 

MARKOV DECISION PROCESS "DUJPO "HFOU &OWJSPONFOU 3FXBSE A t R
t 4UBUF S t R t+1 S t+1 SUPERMARIO WITH R.L 3FXBSE 1FOBMUZ

RETURN 4UBUFWBMVF REINFORCEMENT LEARNING 4UBUF"DUJPOWBMVF

EXPLORATION

https://www.youtube.com/watch?v=8vNxjwt2AqY

EXPLOITATION AND EXPLORATION REINFORCEMENT LEARNING Exploitation Exploration ?

EXPLOITATION REINFORCEMENT LEARNING Exploitation Exploration ?

EXPLORATION REINFORCEMENT LEARNING Exploitation ? Exploration

IMPORTANCE OF EXPLORATION ࣄ੉ 3VTTJBO#MVF ࢓ $VSJPTJUZ  ಽ੉ .VODILJO ࢓
'PPE REINFORCEMENT LEARNING

IMPORTANCE OF EXPLORATION-2 REINFORCEMENT LEARNING ಽ੉ ࣄ੉ ;FSP  FYQMPSBUJPO &YQMPSBUJPO

IMPORTANCE OF EXPLORATION-3 REINFORCEMENT LEARNING ಽ੉ ࣄ੉ 'BJM

GOAL SUPERMARIO WITH R.L

ALL WORLDS AND LEVELS SUPERMARIO WITH R.L env = gym_super_mario_bros.make('SuperMarioBros-<world>-<level>-v<version>')

APPROXIMATE ACTION-VALUE SUPERMARIO WITH R.L

DOUBLE DQN SUPERMARIO WITH R.L JOQVU "DUJPO WBMVF &OW 2/FUXPSL
s’ s 3FQMBZNFNPSZ 2 T B a r (S t , A t , R t+ 1 , S t+ 1 )

1000EPISODES, 3000EPISODES, TRAINING SUPERMARIO WITH R.L FQJTPEFT FQJTPEFT

5000 EPISODE SUPERMARIO WITH R.L FQJTPEFT %BZT

ALPHAGO, DOTA REINFORCEMENT LEARNING

RL REQUIRES A LOT OF TRAINING TIME REINFORCEMENT LEARNING

PRIOR KNOWLEDGE REINFORCEMENT LEARNING

HOW WE CAN ALLOW OUT A.I SYSTEM MAKE TO USE
PRIOR KNOWLEDGE? REINFORCEMENT LEARNING https://ubisafe.org/explore/demeanure-clipart-prior-knowledge/

META LEARNING

2. META-LEARNING META LEARNING Walking the street Get the flag
Go to the target Hit the ball Clear the game Kick the target Find the red ball

REINFORCEMENT LEARNING LEARNING DEXTEROUS IN HAND MANIPULATION

REINFORCEMENT LEARNING META-RL IN ROBOT

ઁоৈ۞࠙җ૓૞ա־Ҋर঻؍ ղ੉ঠӝ

R.L REINFORCEMENT LEARNING Meta Reinforcement Learning Exploration Reinforcement Learning

https://www.facultyfocus.com/articles/teaching-and-learning/wheres-the-curiosity/

WONSEOK JUNG ߊݺо ݅ചо җ೟੗ Ѥ୷о оࣻ Ѻైӝࢶࣻ ੗زର٣੗੉ց ઁಿ٣੗੉ց
੘ҋо Է ఋై੉झ౟ ࣿ૘਍৔ ஠ۨ੉ࢲ झ௢ఠ׮੉ߡ ੘о ಁ࣌٣੗੉ց ೾झ౟ۨ੉ց Үਭо ҃৔ஶࢸఢ౟ ؘ੉ఠࢎ੉঱౭झ౟ ੋҕ૑מোҳਗ Үਭо ੋҕ૑מূ૑פয ۽ࠈੋҕ૑מোҳਗ

https://news.joins.com/article/11522250

https://en.paperblog.com/mcdonald-s-a-glimpse-behind-the-making-off-burgers-502103/

https://medium.com/old-media/why-i-started-reading-a-lot-of-books-8e948aecf5fe

https://m.tmall.com/mblist/de_Abr0pFEQpnxc4QwZUVeDgg==.html

https://thoughtcatalog.com/leigh-wright/2014/12/7-incredible-books-on-exploration-that-will-fuel-your-wanderlust/

https://www.pinterest.ch/pin/506655026820432199/

https://www.slideshare.net/MAAgcunningham/jmm-mini4-sydney-opera-house-part-2

https://www.facultyfocus.com/articles/teaching-and-learning/wheres-the-curiosity/

https://www.rit.edu/science/sms

Ѥ୷о

ߕ٘ࢤഝ

3FTDVFTDVCBEJWFS

2018 QuantNet Rankings of Best Financial Engineering Programs Ranked by
QuantNet in Dec 2017. The most comprehensive 2018 ranking of best Financial Engineering (MFE), Mathematical Finance programs in North America. The 2018 QuantNet ranking of Financial Engineering/Quantitative Finance masters programs in North America provides detailed information on placement and admission statistics from top programs the region, making it uniquely valuable to the quant finance community at large. The 2018 QuantNet rankings are best positioned to help prospective applicants decide where to apply and enroll in those master quantitative programs. 2018 Financial Engineering Programs Rankings Methodology Share your opinion about the 2018 QuantNet ranking. Rank Program Total score Peer assessment score Employment rate at graduation Employment rate 3 months after graduation Average starting base salary (not including bonus) Average GRE Quant of admitted students Tuition Class Size 1 Baruch College, City University of New York Financial Engineering New York, NY 100 4.1 94% 100% $110,000 169.5 $40,980 (non- resident), $27,675 (resident) 36FT, 4PT 1 University of California, Berkeley Financial Engineering Berkeley, CA 100 4.2 84% 96% $108,286 168 $68,725 68FT 3 Carnegie Mellon University Computational Finance Pittsburgh, PA 98 4.3 68% 85% $100,939 169 $83,100 97FT 3 Columbia University Financial Engineering New York, NY 98 3.9 97% 100% $93,000 169 $69,696 99FT

࢜۽਍ҾӘૐ

഑द࠺तೠ੼ਸ଺ਵ࣑աਃ

EXPLOITATION REINFORCEMENT LEARNING Exploitation Exploration ?

REINFORCEMENT LEARNING META-RL IN ROBOT

https://post.naver.com/viewer/postView.nhn?volumeNo=11968470&memberNo=8254994

https://www.intuitionology.com/intuition-ever-lead-wrong-path/

https://www.bodpress.com/1007/alphago-beginner-s-guide/

যځೠݾ಴ܳ׳ࢿೞӝਤ೧ࢲח

Ӓݾ಴ܳ׳ࢿೞӝਤೠৈ۞о૑೐۽ࣁझоઓ੤ೠ׮

tޙઁо৵ߊࢤೡө u

tযڃߑߨਸࢎਊೞݶ੉੹ࠁ׮જই૕ࣻ੓ਸө u

੉૕ޙী׹ਸೞӝਤ೧ࠁా݆਷ҕࠗоਃҳػ׮

https://angel.co/ai-build/jobs

Meta Reinforcement Learning Exploration

ةࢲܳాೠр੽҃೷

૒੽҃೷

WONSEOK JUNG ߊݺо ݅ചо җ೟੗ Ѥ୷о оࣻ Ѻైӝࢶࣻ ੗زର٣੗੉ց ઁಿ٣੗੉ց
੘ҋо Է ఋై੉झ౟ ࣿ૘਍৔ ஠ۨ੉ࢲ झ௢ఠ׮੉ߡ ੘о ಁ࣌٣੗੉ց ೾झ౟ۨ੉ց Үਭо ҃৔ஶࢸఢ౟ ؘ੉ఠࢎ੉঱౭झ౟ ੋҕ૑מোҳਗ Үਭо ੋҕ૑מূ૑פয ۽ࠈੋҕ૑מোҳਗ

ҾӘೠѪ੉ࢤӟ׮ݶ   Ҋ޹ೞҊ   ಽযࠁ۰Ҋ֢۱೧ࠁݶ  যڄөਃ

ೞҊर਷ োҳ઱ઁо ੓׮ݶ ־ҳٚ૑ োҳपਸ ٜ݅ ࣻ ੓Ҋ ੤߀য ࠁ੉ח
োҳपਸ ଺ও׮ݶ ־ҳա োҳपী ଵৈೡ ࣻ ੓ח ݽف ݽৈ ೣԋ োҳೞח ࣁ࢚ী হ؍ ױ ೞա੄ ৌܽ োҳࣗ э੉ೡࣻب੓যਃ

п੗੄ӡਸ଺ই  ৈ۞࠙݅੄੉ঠӝٜܳ݅যࠁݶযڄөਃ

੷ب҅ࣘ&YQMPSBUJPOೞҊ ب੹ೡ҅ദ੉ীਃ

(JUIVC IUUQTHJUIVCDPNXPOTFPLKVOH 'BDFCPPL IUUQTXXXGBDFCPPLDPNXTKVOH #MPH IUUQTXPOTFPLKVOHHJUIVCJP ਋ܻژࠊਃ

꿈꾸는 Agent

꿈꾸는 Agent

More Decks by Wonseok Jung

Other Decks in Research

Featured

Transcript