Cooperative Deep Multi-Agent Reinforcement Learning

8^th September 2021, 11:00
Bei Peng

Abstract

Many real-world learning problems involve multiple agents acting and interacting in the same environment to achieve some common goal, which can be naturally modeled as cooperative multi-agent systems. In this talk I will first overview some of the key challenges in cooperative multi-agent reinforcement learning. I will then describe the problem setting we focus on and the training paradigm we usually use to learn in such settings. One critical challenge in this setting is how to represent and learn the complex joint value functions. I will talk about two deep multi-agent reinforcement learning algorithms we developed recently to address this challenge. Finally, I will present Multi-Agent MuJoCo, a new comprehensive benchmark suite that we developed, based on the popular single-agent MuJoCo benchmark, to allow the study of decentralised continuous control. We believe it can potentially stimulate more progress in continuous multi-agent reinforcement learning.

Biography

Bei Peng will join University of Liverpool as a Lecturer in Artificial Intelligence in the Department of Computer Science starting this September. Her research focuses mainly on deep reinforcement learning, multi-agent systems, interactive machine learning, and curriculum learning. She is currently a Postdoctoral Researcher in the Whiteson Research Lab at the University of Oxford, working with Shimon Whiteson. She is also a Non-Stipendiary Lecturer in Computer Science at St Catherine's College at the University of Oxford. Bei obtained her Ph.D. in the Intelligent Robot Learning Lab at the Washington State University in 2018, supervised by Matthew E. Taylor.