FedECS: Client Selection for Optimizing Computing Energy in Federated Learning
This paper proves the effect of Federated Learning (FL) on lowering computing energy consumption and proposes an FL algorithm for the energy-efficiency-ratio-based client selection (FedECS) and reduces the total energy consumption of FL through the trade-off of iteration rounds and energy consumption per round.
Abstract
With the rapid development of artificial intelligence, computing energy consumption has become a research focus. Due to privacy protection, it is difficult to exchange data between participants, which leads to repeated training of the same model for different clients. This paper aims to reduce the computing energy consumption caused by repeated training of the same model. This paper first proves the effect of Federated Learning (FL) on lowering computing energy consumption. Then considering the client computing architecture, this paper proposes an FL algorithm for the energy-efficiency-ratio-based client selection (FedECS). The proposed algorithm reduces the total energy consumption of FL through the trade-off of iteration rounds and energy consumption per round. In addition, this paper performs extensive simulations, including different data partitions and computing architecture distributions. Experiments show that compared with the FedAvg algorithm, the proposed algorithm can reduce energy consumption by 17.2% ∼ 48.2%.