FedPM: Federated Learning Using Second-order Optimization with Preconditioned Mixing of Local Parameters

Hiro Ishii, Kenta Niwa, Hiroshi Sawada +3

We propose Federated Preconditioned Mixing (FedPM), a novel Federated Learning (FL) method that leverages second-order optimization. Prior methods--such as LocalNewton, LTDA, and F…