BLOCKCHAIN-BASED PRIVACY-PRESERVING AND SECURE FEDERATED LEARNING FRAMEWORK
Federated Learning (FL) allows clients to safely communicate gradients calculated on their local data with the server, eliminating the need for them to reveal their private local datasets. In classical FL, the server may utilize its dominating position to deduce sensitive information from the clients' shared gradients during the model aggregation phase. Malicious clients may also send malicious and counterfeit gradients during model training. Such conduct reduces the utility and dependability of trained models in addition to jeopardizing the integrity of the global model. This paper suggests a Block chain-based Privacy-preserving and Secure Federated Learning (BPS-FL) system that uses threshold homomorphic encryption to safeguard the local gradients of clients in order to successfully solve such privacy and security assault challenges. To prevent malicious gradient attacks, we create a Byzantine-robust aggregation protocol for BPS-FL to enable cipher-text level safe model aggregation. Additionally, our usage of a block chain as the basic distributed architecture for documenting all learning processes ensures the data's immutability and traceability. Our thorough security study and numerical assessment show that BPS-FL can successfully fight off poisoning assaults and meets privacy criteria.