Events Calendar

Reinforcement Learning and Markov Chain Computations, Part I

Speaker: Prof. Vivek Borkar, Tata Institute of Fundamental Research (TIFR), Mumbai, India Part I: Tuesday, Oct. 6, 2-3PM, HED 116 Part II: Thursday, Oct. 8, 2-3PM, EEB 248This two part series shall cover an introduction to reinforcement learning and stochastic approximations, and its application to Markov Chain computations.Part I (Tuesday, Oct. 6) shall highlight the main strands in the reinforcement learning based approaches to approximate dynamic programming for Markov decision processes. In particular, connections to numerical methods for MDPs and convergence issues will be discussed.Part II (Thursday, Oct. 8) will present a novel potential application of reinforcement learning algorithms, viz., for certain matrix computations. It will be argued that these present a hybrid scheme situated between pure Monte Carlo and pure numerical iterative schemes. Various trade-offs and acceleration techniques will be discussed.Speaker Bio: Vivek Borkar is a Professor in the School of Technology and Computer Science at the Tata Institute of Fundamental Research (TIFR), Mumbai, where he has been for the last decade. He was formerly Dean of the same school. Prior to TIFR, he was a Professor in the Computer Science and Automation department of the Indian Institute of Science, Bangalore. He received his Ph.D. from University of California, Berkeley in EECS in 1979. He is well-known for his work in many areas including stochastic processes, mathematical control, game theory and learning. He is the author of several books including a recent book on Stochastic approximations: A Dynamical Systems Viewpoint.Host: Prof. Rahul Jain, 213-740-2246, rahul.jain@usc.edu.

Location: Hedco Pertroleum and Chemical Engineering Building (HED) - 116

Audiences: Everyone Is Invited

Contact: Georgia Lum