Seminar: Machine Learning Seminar

ECE Women Community

Reinforcement Learning under Exogenous Uncertainty

Date: March,15,2023 Start Time: 10:30 - 11:30
Location: 1061, Meyer Building
Add to:
Lecturer: Esther Derman
Affiliations: The Andrew and Erna Viterbi Faculty of Electrical & Computer Engineering

In recent decades, the increasing number of tasks to ensure a better life quality has necessitated going beyond human limits. Reinforcement learning (RL), a branch of artificial intelligence (AI), provides a practical framework for making decisions in a dynamic and evolving world. Yet, critical barriers remain until RL can broadly be deployed. Indeed, to ensure tractability, algorithmic methods often rely on unrealistic assumptions such as a known/fixed environment or immediate execution of an agent’s decision.

In this talk, I will elaborate on these assumptions and propose new methods to overcome them efficiently. More precisely, if time permits, I will cover the following challenges:

How to build online uncertainty sets for less conservative robust RL?

How to avoid the polynomial time complexity of robust Bellman updates so as to scale robust RL to large environments?

How to account for delayed action execution without resorting to the naive, unscalable state-augmentation method?

* Esther Derman is a Ph.D. candidate at the EE Department at the Technion under the supervision of Prof. Shie Mannor.

 

All Seminars
Skip to content