Reinforcement Learning: Where Machine Learning Meets Stochastic Processes

Abstract:

Recently Deep Reinforcement Learning has become a hot topic in the machine learning community. Reinforcement learning is being used to control robots, make sequential decisions in complex simulations, and beat grand master's in classic games such as go and chess. 
Reinforcement Learning (RL) builds on top of Markov Decision Processes (MDPS), studied extensively in the 1960-90s. MDPs are discrete-time optimal stochastic control problems. The main difference between RL and MDPs is that in RL the state transition function and reward distributions are unknown, the state space is huge, and it is possible to inexpensively run vast numbers of simulated trajectories. 
I worked on MDPs in the 1980s as a PhD student and afterwards. After the 1980s I moved on to other subjects, but over the past nine months I've become interested in RL. In the first half of this talk, after defining the RL problem in the context of MDPs, I will outline a classic convergence proof in RL. I will then describe a deep reinforcement learning problem we are currently working on. 
RL, and more generally Machine Learning, is full of intriguing and impactful probability theory problems. One of the goals of the talk will be to pique the interest of our talented probabilists at NYU Shanghai. 

NYU Shanghai STEM seminar series is a weekly seminar series on every Wednesday, starting from 12th October 2016. 
Please see below tentative schedule of STEM seminar series in 2017 Spring Semester.

• Feb. 8: NO SEMINAR
• Feb. 15: Li Li, Associate Professor of Neural Science and Psychology
• Feb. 22: Xiao-Jing Wang, Global Professor of Neural Science  
• Mar. 1: Hanghui Chen, Assistant Professor of Physics
• Mar. 8: Vladas Sidoravicius, Professor of Mathematics
• Mar. 15: Keith Ross, Professor of Engineering and Computer Science
• Mar. 22: Olivier Marin, Associate Professor of Computer Science
• Mar. 29: Xinying Cai, Assistant Professor of Neural and Cognitive Sciences
• Apr. 5: Spring Recess, NO SEMINAR
• Apr. 12: Jungseog Kong, Assistant Professor of Biology
• Apr. 19: Gang Fang, Assistant Professor of Biology
• Apr. 26: Pierre Michael Tarres, Visiting Professor of Mathematics
• May 3: Jeff Erlich, Assistant Professor of Neural and Cognitive Sciences
• May 10: Laurent Mertz, Visiting Assistant Professor of Mathematics
• May 17: NO SEMINAR