Monday, November 27, 2023 - 4:00pm
Event Calendar Category
LIDS Seminar Series
Building and Room Number
We begin by studying the problem of learning a linear system model from the observations of M clients. The catch: Each client is observing data from a different dynamical system. This work addresses the question of how multiple clients collaboratively learn dynamical models in the presence of system and environmental heterogeneity. We pose this problem as a federated learning problem and characterize the tension between achievable performance and system heterogeneity. Our federated sample complexity result provides a constant factor improvement over the single agent setting in the low heterogeneity regime. We then show how clustering can improve these results. Finally, we will extend this framework to the setting of federated optimal control design using policy gradients.
James Anderson is an assistant professor at Columbia University in the Department of Electrical Engineering and a member of the Data Science Institute. Prior to joining Columbia he was a senior postdoctoral scholar at the California Institute of Technology in the Computing and Mathematical Sciences Department. From 2012-2016 he held a junior research fellowship in Engineering Science at St John’s College, Oxford. He received his DPhil from the University of Oxford in 2012.