Speaker |
David Siska |
|---|---|
Affiliation |
University of Edinburgh |
Date |
Wednesday, 19 November 2025 |
Time |
12:30-13:30 |
Location |
Ground floor lecture theatre, Sainsbury Wellcome Center, 25 Howland St, W1T 4JG |
Link |
https://ucl.zoom.us/j/99748820264 |
Event series |
Jump Trading/ELLIS CSML Seminar Series |
Abstract |
Policy gradient methods have been used to develop efficient reinforcement learning (RL) algorithms over the last decade. This talk will review recent results on the convergence of policy-gradient-type methods for RL in general state and action spaces, focusing on the mirror descent method with and without function approximation, its connection to the natural policy gradient (NPG), and the role of convexity (or lack thereof) in the objective function. |
Biography |
I am a Reader in the School of Mathematics with research interests spanning the theory and applications of stochastic systems. My work focuses on the mathematics of reinforcement learning, numerical methods for stochastic control, and stochastic analysis, including partial differential equations and McKean–Vlasov stochastic differential equations. I am also broadly interested in applications of these methodologies in engineering, economics, finance, and insurance. |