Speaker: Nouman Khan (University of Michigan, Electrical Engineering and Computer Science)
Title: Foundations for Cooperative Multi-Agent Systems and Networks: Decentralized Decision Making
Date: April 5, 2024 (Friday)
Time: 15:00 – 16:00
Place: Zoom
This is an online seminar. To obtain event details please send a message to department.
Abstract: The availability of big data and the rapid growth in storage, sensing, communication, and computing technologies are disrupting modern engineering systems and networks. There is a pressing demand for data-driven control algorithms with reinforcement learning (RL) having taken the center stage—largely attributed to its successes in the cyber world such as in game playing, recommendation systems, and finance. Despite these feats, today’s state-of-the-art RL algorithms are not yet optimal, scalable, or efficient enough for reliable deployments in large-scale systems and networks, especially those that are safety-critical or resource-constrained. This barrier to applying modern-day RL is due to three key phenomena that are unique to large-scale systems and networks and that need adequate consideration, namely: i) multi-agent interactions, ii) complex information structures (due to partial observability or information asymmetry), and iii) joint constraints. In this talk, I will discuss my research thus far on the study of such systems, considering the mentioned phenomena. The highlight of the talk will be the framework of cooperative “Multi-Agent Constrained POMDP” (MA-C-POMDP) and a fundamental strong duality result that was established during my Ph.D.—serving as the groundwork for a principled design of primal-dual type Multi-Agent Reinforcement Learning (MARL) algorithms. I will also briefly share one important research direction that I am excited to pursue in the study of large-scale systems and networks.
Bio: Nouman Khan is a final year Ph.D. candidate in the Electrical Engineering and Computer Science (EECS) Department at the University of Michigan (UoM), Ann Arbor, working with Professor Vijay Subramanian. He received his B.S. degree in Electrical and Electronic Engineering from the GIK Institute of Engineering Sciences and Technology (GIKI) Pakistan in 2014, and his M.S. degree in ECE from UoM in 2019. In summers of 2022 and 2023, he was an Applied Scientist intern at “Amazon Search Science and AI” where he helped productionize deep contextual bandits and counterfactual learning pipelines. Between 2014 and 2017, he worked as a Controls and Instrumentation Engineer at MOL Group and PGNiG. He is the recipient of UoM’s Rackham International Student Fellowship (2023) and the Towner Prize for Distinguished Academic Achievement (2024). His research focuses on developing theoretical foundations and determining performance limits for scalable and efficient solutions in the study of multi-agent systems, networks, and network phenomena.