Escaping Saddle Points in Heterogeneous Federated Learning via Distributed SGD with Communication Compression

Open in new window