Multiagent learning is a fundamental component of multiagent systems both from scientific and engineering perspectives. From a scientific perspective, studying the interactions among learning agents provides insight into many social phenomena, from game theory to commodities trading to resource allocation problems. From an engineering perspective, learning agents provide a conceptually proven approach to distributed control problems such as load balancing, sensor networks, multi-robot coordination, and air traffic management (Tuyls and Tumer, 2013, Chapter 10 Multiagent Learning, 423–483, edited by G. Weiss, Multiagent Systems, 2nd edition, 2013, MIT Press).
This section focuses on techniques drawn primarily from two disciplines, AI and game theory, although those in turn borrow from a variety of disciplines, including control theory, statistics, psychology, and biology, to name a few. This session will discuss various subtle aspects (e.g., what constitutes learning?) of learning in multiagent systems and then discuss representative theories in this area: model-based learning, rational learning (Bayesian learning), no-regret learning, reinforcement learning (RL) in multiagent extensions, and evolutionary learning.
Updated May 04 2018 by FST Course Production Staff