Energy Efficiency Optimized Federated Learning Algorithm in Wireless Networks
Liangduo Shen, Tiankui Zhang · 2023
Learning the trade-off between training accuracy and energy loss in performing is a federated learning task. Firstly, the ratio of system energy expenditure for communication and computation to federated learning accuracy is considered as an energy efficiency optimization problem, for the joint optimization of global parameter aggregation iterations and local training iterations. Further, through mathematical analysis, the convergence of Lipschitz smooth loss functions is proven, forming the basis to obtain a relationship between local training iterations and the loss function, and deriving a federated learning accuracy expression with local training iterations as a variable. Using the SCA (Successive Convex Approximation) method, the proposed non-convex problem is transformed into a convex one, and the corresponding energy-efficient federated learning algorithm is proposed. Simulation results indicate that the proposed algorithm, when deployed on federated learning systems for training, results in less communication and computational system expenditure than other comparative algorithms, while achieving the constrained accuracy conditions.