Agnostic Federated Learning

TL;DR

This paper introduces agnostic federated learning, a framework that optimizes models for any mixture of client data distributions, promoting fairness and robustness across diverse and shifting data environments.

Contribution

It proposes a novel agnostic federated learning framework, provides theoretical guarantees, and develops a fast stochastic algorithm with proven convergence, demonstrated empirically on multiple datasets.

Findings

The framework achieves improved fairness across clients.

The algorithm converges efficiently under convex loss assumptions.

Empirical results show benefits over traditional federated learning methods.

Abstract

A key learning scenario in large-scale applications is that of federated learning, where a centralized model is trained based on data originating from a large number of clients. We argue that, with the existing training and inference, federated models can be biased towards different clients. Instead, we propose a new framework of agnostic federated learning, where the centralized model is optimized for any target distribution formed by a mixture of the client distributions. We further show that this framework naturally yields a notion of fairness. We present data-dependent Rademacher complexity guarantees for learning with this objective, which guide the definition of an algorithm for agnostic federated learning. We also give a fast stochastic optimization algorithm for solving the corresponding optimization problem, for which we prove convergence bounds, assuming a convex loss function and hypothesis set. We further empirically demonstrate the benefits of our approach in several datasets. Beyond federated learning, our framework and algorithm can be of interest to other learning scenarios such as cloud computing, domain adaptation, drifting, and other contexts where the training and test distributions do not coincide.