DECENTRALIZED RATIONAL FEDERATED LEARNING SCHEME BASED ON FULLY HOMOMORPHIC ENCRYPTION

Aimed at the problems of model parameters and data privacy security in federated learning process, a decentralized federated learning scheme based on fully homomorphic encryption is proposed based on blockchain smart contract and Pedersen promise technology. The game theory was used to design incentive mechanisms for federated learning, and the utility function was used to motivate cooperative nodes to actively participate in the model training, so as to reduce the communication cost of uploading model parameter verification in the process of federated learning. In addition, in order to verify the effectiveness of the scheme, this paper respectively built the prototype of the accuracy, loss, time and communication consumption of node model training. Through simulation of training nodes with different traditional and rational models, the experiment verifies that the proposed federated learning scheme effectively reduces the communication cost of task training process model verification.