Mean Field Control (MFC) as an Approximation for Cooperative Multi Agent Reinforcement Learning (MARL)

# 129


Mean field control (MFC) is an effective way to mitigate the curse of dimensionality of cooperative multi-agent reinforcement learning (MARL) problems. In this work, we will show that MFC is indeed a good approximation to MARL in variety of setups, including heterogenous agents, non-uniform interaction, and in the presence of constraints. Further, the approach of MFC allows for decentralized execution, which will also be discussed.

Vaneet Agarwal, Purdue University

Vaneet Aggarwal is Associate Professor with the School of Industrial Engineering and the School of Electrical and Computer Engineering, at Purdue University since Jan 2015. He was with AT&T Labs - Research (also called AT&T Shannon Labs), and worked in the Artificial Intelligence and Communications Research Group, Wireless Network Technology Research Group, and Service Quality Management Research Group at certain times during 2010-2014. He did his Ph.D. from the Electrical Engineering Department at Princeton. He did his undergraduate in Electrical Engineering at Indian Institute of Technology Kanpur, India. His research interests are Machine Learning, Networking, and Quantum Computing.