Applied Mathematics Colloquium with Sebastian Jaimungal:Time Consistent Risk-Aware Reinforcement Learning

Speaker: Sebastian Jaimungal, University of Toronto

Title: Time Consistent Risk-Aware Reinforcement Learning

Abstract: Reinforcement learning (RL) is a model agnostic approach for “learning” optimal strategies where an agent is faced with maximizing or minimizing an objective functional over a class of strategies that are, in general, stochastic processes. Traditionally, in RL, the objective functionals are expectations of total running rewards/costs – and ignores risks. While in the extant literature there are various ad hoc approaches for incorporating risk, these approaches typically lead to what are known as time inconsistent solutions. In this talk, I will provide an overview of RL, the time inconsistency issue, and introduce some new approaches to pose and solve time consistent versions of risk-aware RL problems with applications to financial modeling.

[ based on joint works with Alvaro Cartea, Anthony Coache, Silvana Pesenti, Yuri Saporito, and Rodrigo Targino ]

 

 

Applied Mathematics Colloquia