Federated Learning with Heterogeneous Architectures using Graph HyperNetworks

Or Litany, Haggai Maron, David Acuna, Jan Kautz, Gal Chechik, Sanja Fidler

Keywords: federated learning, graph neural networks, hypernetworks

Abstract: Standard Federated Learning (FL) techniques are limited to clients with identical network architectures. As a result, inter-organizational collaboration is severely restricted when both data privacy and architectural proprietary are required. In this work, we propose a new FL framework that removes this limitation by adopting a graph hypernetwork as a shared knowledge aggregator. A property of the graph hyper network is that it can adapt to various computational graphs, thereby allowing meaningful parameter sharing across models. Unlike existing solutions, our framework makes no use of external data and does not require clients to disclose their model architecture. Compared with distillation-based and non-graph hypernetwork baselines, our method performs notably better on standard benchmarks. We additionally show encouraging generalization performance to unseen architectures.

One-sentence Summary: A graph-hypernetwork-based solution for federated learning where clients have different neural architectures