A Byzantine-Robust Method for Decentralized Federated Learning through Gradient Filtering and Weight Transformation

Decentralized federated learning (DFL) is an emerging paradigm that enables collaborative model training across multiple clients without the need for a central coordinator. The decentralized structure enhances privacy and scalability but is vulnerable to Byzantine attacks, where malicious clients can disrupt the training process by sending incorrect updates. To address this challenge, a novel Byzantine-robust decentralized federated learning algorithm is proposed, integrating gradient filtering and weight matrix transformation techniques. The proposed method enhances robustness by filtering out potentially malicious gradients and dynamically adjusting the communication topology based on trust levels between nodes. Experimental results demonstrate that the approach improves the model's resilience to Byzantine attacks compared to traditional federated learning methods.