AIR Distinguished Speaker Series: Using Function Approximation to Solve Large Scale Planning Problems in MDPs
- Starts: 11:00 am on Monday, December 6, 2021
- Ends: 12:00 pm on Monday, December 6, 2021
Csaba Szepesvari, Professor of Computing Science, University of Alberta. Abstract: Bellman and his co-workers already in the 1960s have used linear function approximation (linearly combined basis functions) to solve planning problems in Markov Decision Processes. Their hope was that this approach can scale to large problems. In this talk, starting from the assumption that the basis functions are such that the optimal value function is representable with them, we look at some recent results that provide interesting answers to this ancient question.