Efficient and Verifiable Federated Learning Based on Blockchain

Federated learning (FL) breaks the data island crisis and enables a large number of users to collaboratively train a global model. The existing federated learning schemes have low efficiency in secure aggregation, and users have to trust a single server. To address the above issues, we propose an efficient and verifiable federated learning scheme based on blockchain. First, blockchain and smart contract are used to implement multi-user collaborative training without a trusted center. Second, double masking and homogeneous linear recursion based secret sharing are used to achieve the goal of protecting user privacy and improving aggregation efficiency. Finally, we verify the correctness of the aggregation results using vector homomorphic hashing and smart contract. Security analysis proves the confidentiality and verifiability of the scheme. Simulation results show that the efficiency is improved by about $47 \%$ compared with the existing scheme.