reinfocement learning

© 2016. SNU CSE Biointelligence Lab., http://bi.snu.ac.kr1

Introduction of Reinforcement Learn-ing

곽동현

서울대학교 바이오지능 연구실


Background• 기존의 강화학습 (Reinforcement Learning) 에서 Q func-

tion 을 DNN 혹은 CNN 으로 근사하여 문제를 해결하는 시도가 최근 Google DeepMind 를 필두로 활발히 연구가 되고 있다 .

• 최근 연구에서는 Atari 2600, 바둑을 인간보다 더 잘 플레이하는 수준의 경이적인 성과를 보이고 있으며 , 나아가 3D 게임이나 로봇 컨트롤 문제에도 적용되고 있다 .


What is AI? ML?

https://www.linkedin.com/pulse/deep-dive-venture-landscape-ai-ajit-nazre-rahul-garg-nazre


Various Field with ML

https://www.linkedin.com/pulse/how-exceed-your-goals-2016-dr-travis-bradberry-1





Various Algorithm in ML


Function Approximation

http://arxiv.org/pdf/1411.4555.pdf https://people.mpi-inf.mpg.de/~kkim/supres/supres.htm

http://arxiv.org/pdf/1411.4555.pdf

http://arxiv.org/pdf/1411.4555.pdf

https://people.mpi-inf.mpg.de/~kkim/supres/supres.htm

https://people.mpi-inf.mpg.de/~kkim/supres/supres.htm


What is Deep Learning?


Machine Learning• Supervised Learning :

y = f(x)

• Unsupervised Learning : x ~ p(x) , x = f(x)

• Reinforcement Learning : ??


Agent-Environment Interaction

• Objective : Maximize the expected sum of future rewards

• Algorithms1) Planning : Dynamic Programming Based2) Reinforcement Learning : Machine Learning Based


Example of Super-vised Learning


Polynomial Curve Fitting

Microsoft Excel 2007 의 추세선


Example of Un-upervised Learning


Clustering

http://www.frankichamaki.com/data-driven-market-segmentation-more-effective-market-ing-to-segments-using-ai/


Example of Rein-forcement Learning


Videos• A crawling robot: a Q-learning examplehttps://www.youtube.com/watch?v=2iNrJx6IDEo

• Deep Reinforcement Learning for Robotic Ma-nipulation

https://youtu.be/ZhsEKTo7V04?t=1m27s

https://youtu.be/UU4ULmd59sg?t=47s






THANK YOU

reinfocement learning

Technology