Blockchain-based Approaches for Secure Federated Learning
Lucas Airam C. de Souza, Gustavo F. Camilo, Gabriel Antonio F. Rebello, Lucas C. B. Guimarães, Miguel Elias M. Campista, Luís Henrique M. K. Costa · 2024
In federated learning, different clients can contribute to the learning process without disclosing private information. Nevertheless, other security issues do still exist. Malicious users can delay the convergence or even degrade the final global model by injecting untrained models. This paper integrates blockchain technology with federated learning and evaluates the performance based on small committees. Based on real experiments, we show that registering all model updates in blockchain introduces system auditability without impacting the federated learning performance.