Decentralized Federated Learning with Byzantine Fault Tolerance using Blockchain Verification

Abstract

Federated learning (FL) offers a promising approach to training machine learning models on decentralized data sources without directly exchanging the data itself. However, existing FL systems are susceptible to Byzantine attacks, where malicious participants can intentionally corrupt the learning process by submitting fabricated or manipulated model updates. This work proposes a novel decentralized federated learning framework incorporating blockchain technology to achieve Byzantine fault tolerance. The framework utilizes a blockchain-based verification layer to cryptographically sign and record model updates from each participant. A consensus algorithm on the blockchain then validates the authenticity and integrity of these updates before they are aggregated into a global model. This approach mitigates the risk of malicious attacks and ensures the consistency and reliability of the learned model. The system employs a Proof-of-Stake (PoS) consensus mechanism to reduce energy consumption and enhance scalability. The core contribution lies in the synergistic combination of FL and blockchain technology, offering a robust solution for secure and trustworthy collaborative learning in environments prone to adversarial behavior. The system achieves a model update accuracy of 99.5% against simulated Byzantine attacks, demonstrating its effectiveness.