Federated Multi-Task Learning

TL;DR

This paper introduces MOCHA, a systems-aware optimization method for federated multi-task learning that addresses communication costs, stragglers, and fault tolerance, demonstrating significant speedups in real-world simulations.

Contribution

It presents the first theory and method that explicitly handle practical systems issues in federated multi-task learning, improving efficiency and robustness.

Findings

MOCHA achieves significant speedups over existing methods.

Theoretical analysis considers communication, stragglers, and fault tolerance.

Demonstrated effectiveness on real-world federated datasets.

Abstract

Federated learning poses new statistical and systems challenges in training machine learning models over distributed networks of devices. In this work, we show that multi-task learning is naturally suited to handle the statistical challenges of this setting, and propose a novel systems-aware optimization method, MOCHA, that is robust to practical systems issues. Our method and theory for the first time consider issues of high communication cost, stragglers, and fault tolerance for distributed multi-task learning. The resulting method achieves significant speedups compared to alternatives in the federated setting, as we demonstrate through simulations on real-world federated datasets.