Bayesian Nonparametric Federated Learning of Neural Networks
This addresses federated learning challenges for distributed data servers, but appears incremental as it builds on existing methods without claiming broad SOTA.
The paper tackled the problem of federated learning where data cannot be pooled by developing a Bayesian nonparametric framework to synthesize a global neural network from local weights without extra supervision or data exchange, achieving efficacy in simulations on image classification datasets.
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.