Présentation sur YouTube

This talk will discuss some of what I feel to be key issues in dealing with reinforcement learning in non-Markovian environments, drawing parallels with some classical ideas in stochastic control. This will be used to motivate a particular objective, viz., approximation of certain conditional laws, which is very much in the tradition of control of POMDPs using belief states.

Biography: Vivek Borkar is currently an emeritus faculty at the Indian Institute of Technology Bombay, Mumbai. He studied electrical engineering at IIT Bombay and subsequently did his M.S. and Ph.D. in resp. Case Western Reserve Uni. and Uni. of California at Berkeley. His interests are in stochastic control and optimization, covering theory, algorithms and applications.