Bayesian Nonparametric Federated Learning of Neural Networks

TL;DR

This paper introduces a Bayesian nonparametric framework for federated learning of neural networks, enabling the synthesis of a more expressive global model with minimal communication and without data pooling.

Contribution

It presents a novel Bayesian nonparametric approach that models local neural network weights for federated learning, reducing communication rounds and avoiding data pooling.

Findings

Effective global model synthesis with single communication round

Applicable to image classification datasets in federated settings

Outperforms traditional federated learning methods

Abstract

In federated learning problems, data is scattered across different servers and exchanging or pooling it is often impractical or prohibited. We develop a Bayesian nonparametric framework for federated learning with neural networks. Each data server is assumed to provide local neural network weights, which are modeled through our framework. We then develop an inference approach that allows us to synthesize a more expressive global network without additional supervision, data pooling and with as few as a single communication round. We then demonstrate the efficacy of our approach on federated learning problems simulated from two popular image classification datasets.