Robust federated learning defends clients from faulty neighbors
Robust Decentralized Personalized Federated Learning via Prediction-Constrained Neighborhood Collaboration
Machine LearningCryptography and SecurityDistributed, Parallel, and Cluster Computing
Summary
Federated learning lets many devices train their own personalized AI models while sharing information. However, some devices may send harmful, false updates that can harm the learning process. The authors propose a new method called R-DPFL that helps each device check if updates from its neighbors make sense by comparing them to what it thinks should happen based on past updates. This way, the method reduces the bad effects from harmful devices and lets each device learn more effectively without needing everyone to agree exactly.
What this means in practice
- •For mobile app developers: Build more reliable federated learning apps that stay accurate even when some devices send corrupted data updates.
- •For edge computing engineers: Create edge AI systems that maintain personalized learning on devices connected in decentralized networks under adversarial conditions.
Authors
Xiao Ma, Hong Shen, Hui Tian, Wenqi Lyu, Wei Ke
Abstract
This paper proposes a robust decentralized personalized federated learning method R-DPFL, that enables clients to reduce the impact of Byzantine attacks via robust neighborhood direction estimation and history-based update trend prediction, rather than purely aggregating client models as in the existing work. In R-DPFL, each client first computes the current-round model update by aggregating the received neighborhood update vectors. It then predicts what this update should be based on its historical values and local model changes. Finally, R-DPFL computes the difference between these two quantities, adaptively clips this difference, and adds it to the local update. We prove convergence of the learning process through rigorous analysis and show that honest clients maintain stable personalized descent dynamics under Byzantine neighbor perturbations without requiring consensus among neighboring models. Extensive experiments on CIFAR-10 demonstrate that RDPFL consistently outperforms state-of-the-art decentralized and personalized federated learning baselines under heterogeneous and adversarial settings.