A blockchain-assisted federated learning wireless network model

Abstract

A blockchain-assisted federated learning wireless network model is proposed. Edge devices send locally trained model updates to a central server, where smart contracts are used to verify the uploaded parameters and the server aggregates them. During the aggregation phase, a wireless federated learning algorithm is used to adapt to the wireless environment. Finally, the model parameters are returned to the edge devices to start the next round of learning. The combination of federated learning and blockchain protects against untrusted users or parameter poisoning attacks, thereby improving the reliability and security of the system. The optimization problem of computing and communication resource allocation is solved by decomposing the non-convex problem into several sub-problems, which are solved separately using convex optimization and continuous convex approximation methods to minimize time and energy costs. Numerical results are given for the resource optimization problem, and the algorithm is experimentally evaluated on multiple datasets using PyTorch. Numerical and experimental results show that the proposed scheme effectively improves total energy consumption and mitigates the impact of data poisoning attacks on the learning process.

Images