FedSysID: A Federated Approach to Sample-Efficient System Identification

TL;DR

FedSysID introduces a federated learning framework for efficiently identifying linear system models across multiple heterogeneous clients, achieving better sample complexity than single-agent approaches.

Contribution

This work formulates a federated learning approach for system identification with heterogeneous data, providing theoretical sample complexity improvements and a novel meta-algorithm.

Findings

Federated approach improves sample complexity by a constant factor.

Addresses heterogeneity in multi-client system identification.

Proposes FedSysID, leveraging existing federated algorithms.

Abstract

We study 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 heterogeneity. We pose this problem as a federated learning problem and characterize the tension between achievable performance and system heterogeneity. Furthermore, our federated sample complexity result provides a constant factor improvement over the single agent setting. Finally, we describe a meta federated learning algorithm, FedSysID, that leverages existing federated algorithms at the client level.