Federated Learning Based on Blockchain with Improved Proof of Authority Consensus

To address robustness issues in federated learning, such as central server single points of failure and poisoning attacks by malicious nodes, we propose a Federated Learning framework based on blockchain with improved Proof of Authority Consensus (APBFL). We integrate a specialized Proof of Authority consensus to enhance efficiency and incorporate a credibility mechanism to penalize malicious and negative nodes, mitigating poisoning attacks and system negativity. Experimental results demonstrate the effectiveness of APBFL.