FedADMM: A Federated Primal-Dual Algorithm Allowing Partial Participation
This work addresses communication efficiency and heterogeneity in federated learning, but it appears incremental as it builds on existing ADMM methods for a specific optimization setting.
The authors tackled the problem of federated learning with non-convex composite optimization and non-smooth regularizers by proposing FedADMM, a new algorithm that allows partial client participation, and they proved its convergence under a general sampling model.
Federated learning is a framework for distributed optimization that places emphasis on communication efficiency. In particular, it follows a client-server broadcast model and is particularly appealing because of its ability to accommodate heterogeneity in client compute and storage resources, non-i.i.d. data assumptions, and data privacy. Our contribution is to offer a new federated learning algorithm, FedADMM, for solving non-convex composite optimization problems with non-smooth regularizers. We prove converges of FedADMM for the case when not all clients are able to participate in a given communication round under a very general sampling model.