Differential Optimization Federated Incremental Learning Algorithm Based on Blockchain

Federated learning is a hot area of concern in the field privacy protection. There are local model parameters that difficult to integrate, poor timeliness, and training security issues. This paper proposes blockchain-based differential optimization federated incremental algorithm, First, we apply weighted random forest optimize reduce impact adding on accuracy model. Using different ensemble algorithms integrate can improve global At same time, risk data leakage caused by gradient update reduced; then, applied framework timeliness model; finally, phase uploaded blockchain synchronized quickly, which reduces cost storage parameter transmission. The experimental results show stacking each period above 83.5% variance lower than 10−4 for public set. has been improved, have improved.