Technical - Partner Event
AI Seminar – Sriram Ganapathi Subramanian
When
May 14, 2021 - May 14, 2021
12:00 PM – 1:00 PM MST
Where
Online
Online
Abstract: Mean-field theory provides an effective way of scaling multi-agent reinforcement learning algorithms to environments with many agents by abstracting other agents into a virtual mean agent. This talk will give a brief introduction to the recent research field of mean-field reinforcement learning, which is a very practical technique that can be applicable to a set of real-world problems. Yet, there are some limiting assumptions of previous approaches in this area that prevent the wide applicability of these methods. First, all agents in the environment must be homogeneous. Second, the mean-field metric must be fully observable. I will discuss these assumptions and explain some of our research work that relaxes each of these assumptions. Specifically, I will introduce a multi-type mean-field reinforcement learning approach that will relax the first assumption and a partially observable mean-field reinforcement learning approach that will relax the second assumption. Further, I will also provide practical algorithms for both these methods that have similar theoretical guarantees to previous algorithms in this area in addition to stronger empirical performance compared to baselines on a set of large games with many agents.
Presenter Bio: Sriram is a Ph.D. student in the Department of Electrical and Computer Engineering at the University of Waterloo. He is also a postgraduate affiliate at the Vector Institute, Toronto. His primary research interest is in the area of multi-agent systems. Particularly he is interested in the issues of scale, non-stationarity, communication, and sample complexity in multi-agent learning systems. His research is motivated by the field of computational sustainability. His long-term research vision is to make multi-agent learning algorithms applicable to a variety of large-scale real-world problems. Before starting the Ph.D. program, he obtained a Master's in Electrical and Computer Engineering at the University of Waterloo, Canada in 2018 and a Bachelor's in Geomatics from Anna University, India in 2016. He was the recipient of prestigious fellowships such as the MITACS Globalink Research award, MITACS Graduate Fellowship, Pasupalak fellowship in AI, and Vector postgraduate research award. He has also worked as a research intern in Borealis AI - Edmonton and Waterloo labs.
The University of Alberta Artificial Intelligence (AI) Seminar is a weekly meeting where researchers (including students, developers, and professors) interested in AI can share their current research. Presenters include local speakers from the University of Alberta and industry as well as other institutions. The seminars discuss a wide range of topics related in any way to Artificial Intelligence, from foundational theoretical work to innovative applications of AI techniques to new fields and problems are of interest. Learn more at the AI Seminar website and by subscribing to the mailing list!
Looking to build AI capacity? Need a speaker at your event?
