A Robust Aggregation Method Compatible with Differential Privacy in Federated Learning
Abstract
Robustness and privacy preservation are two fundamental challenges in building trustworthy federated learning (FL) systems. Existing FL approaches typically address robustness through robust aggregation algorithms and ensure data privacy via differential privacy (DP) mechanisms. However, these two objectives are often tackled independently, and their integration can lead to mutual interference. To address this issue, we propose FedSTAR (Federated learning with Statistical and Trained feature-enhanced Robust aggregation), a novel robust aggregation method that is compatible with differential privacy. FedSTAR combines multi-dimensional statistical features with deep feature extraction driven by neural networks and incorporates a historical feature enhancement mechanism to improve the detection of malicious updates. This design helps to mitigate the impact of DP-induced noise while enabling robust global model aggregation and preserving data privacy. Extensive experiments on two benchmark datasets demonstrate that FedSTAR significantly enhances both privacy protection and robustness in federated learning, while maintaining competitive task performance. Notably, in complex tasks, FedSTAR reduces the backdoor attack success rate to below 10%, which is 60% lower than that of the state-of-the-art baselines.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.