Towards Efficient Decentralized Federated Learning

Abstract

We focus on the problem of efficiently deploying a federated learning training task in a decentralized setting with multiple aggregators. To that end, we introduce a number of improvements and modifications to the recently proposed IPLS protocol. In particular, we relax its assumption for di-rect communication across participants, using instead indirect communication over a decentralized storage system, effectively turning it into a partially asynchronous protocol. Moreover, we secure it against malicious aggregators (that drop or alter data) by relying on homomorphic cryptographic commitments for efficient verification of aggregation. We implement the modified IPLS protocol and report on its performance and potential bottlenecks. Finally, we identify important next steps for this line of research.