A Novel Framework for Distributed and Collaborative Federated Learning based on Blockchain and Smart Contracts

Abstract

Blockchain and federated learning, as two key technologies for trusted and privacy-preserving collaboration in distributed environments, have been intensively studied in recent years. Federated learning aims to train a centralized global model from decentralized datasets without leaking user privacy, while blockchain helps establish mutual trusts among multiple clients with technical features such as tamper-proof, anonymity, security and traceability, among others. Specially, blockchain-based smart contracts can perform complex logics and behaviors in an efficient and accurate fashion to schedule collaborations. Therefore, leveraging these advantages, this paper proposed a novel framework for enabling distributed and collaborative federated learning based on blockchain and smart contracts, and discussed its major components, i.e., the blockchain-based distributed architecture, the smart-contract-based scheduling, as well as the incentive mechanism design for federated learning. We also discussed the potential application scenarios of our framework, which can be expected to help establish safer, fairer, smarter, and more efficient collaborations for distributed federated learning.