Enhancing Security in Parallel Federated Learning with Sharded Blockchain for Internet of Vehicles

The Internet of Vehicles (IoV) has experienced rapid growth, generating extensive data that can be leveraged for intelligent applications. Federated learning provides a privacy-preserving approach to utilize this data but faces challenges in performing multi-task learning and ensuring security. To address these issues, we propose a secure Federated Learning framework with Sharded Blockchain (FLSB), where the blockchain design aligns with FL requirements to enhance security, robustness, and scalability. FLSB partitions the blockchain into sharded chains to support parallel FL tasks, while a main chain governs task management and security enforcement. To prevent malicious behaviors such as creating fake identities or disrupting consensus, we develop the Proof of List Update (PoLU) consensus on the main chain. For secure FL, we design a committee-based adaptive weighted aggregation algorithm, whcih dynamically adjusts the model weights during aggregation to defend against poisoning attacks. In addition, we introduce the Proof of Task Participation (PoTP) consensus on sharded chains to ensure trustworthy model aggregation and defend against adversarial manipulations. Experimental results on public datasets demonstrate that FLSB effectively balances FL task performance, security, and privacy, making it well-suited for IoV environments.