BEFL: Balancing Energy Consumption in Federated Learning for Mobile Edge IoT

TL;DR

BEFL tackles energy challenges in Federated Learning for Mobile Edge IoT by balancing accuracy, energy usage, and energy fairness across devices. It introduces SLSQP-based communication resource allocation, a cluster-based energy-balancing client selection heuristic, offline imitation learning for pretraining, and online ranking-based RL to optimize client participation. The method defines a multiobjective objective combining variance in energy consumption, total energy, and model accuracy, with a system model that includes per-device training energy, transmission energy, and OFDMA rates. Experiments on CIFAR-10 and MNIST show BEFL improves global accuracy by 1.6%, reduces energy variance by 72.7%, and lowers total energy by 28.2% compared to baselines, demonstrating practical potential for energy-aware FL in MEIoT.

Abstract

Federated Learning (FL) is a privacy-preserving distributed learning paradigm designed to build a highly accurate global model. In Mobile Edge IoT (MEIoT), the training and communication processes can significantly deplete the limited battery resources of devices. Existing research primarily focuses on reducing overall energy consumption, but this may inadvertently create energy consumption imbalances, leading to the premature dropout of energy-sensitive devices.To address these challenges, we propose BEFL, a joint optimization framework aimed at balancing three objectives: enhancing global model accuracy, minimizing total energy consumption, and reducing energy usage disparities among devices. First, taking into account the communication constraints of MEIoT and the heterogeneity of devices, we employed the Sequential Least Squares Programming (SLSQP) algorithm for the rational allocation of communication resources. Based on this, we introduce a heuristic client selection algorithm that combines cluster partitioning with utility-driven approaches to alleviate both the total energy consumption of all devices and the discrepancies in energy usage.Furthermore, we utilize the proposed heuristic client selection algorithm as a template for offline imitation learning during pre-training, while adopting a ranking-based reinforcement learning approach online to further boost training efficiency. Our experiments reveal that BEFL improves global model accuracy by 1.6\%, reduces energy consumption variance by 72.7\%, and lowers total energy consumption by 28.2\% compared to existing methods. The relevant code can be found at \href{URL}{https://github.com/juzehao/BEFL}.

Figures (7)

• Figure 1: The two issues in energy consumption optimization for FL: 1) The impact of device heterogeneity on energy consumption ensitivity. 2) The imbalance in energy consumption among devices.
• Figure 2: BEFL framework overview: 1. The server retrieves hardware information from devices. 2. It simulates energy consumption and latency using energy consumption model. 3. Client selection is facilitated by a heuristic algorithm. 4. Generated state-action pairs are sent to the RL network for pre-training. 5-11. During online training, the RL agent gathers state information, selects clients, conducts local training, aggregates parameters, and updates both main and target networks. After R rounds, the target network is updated with main network parameters based on Q values and rewards.
• Figure 3: Fedavg
• Figure 4: FlashRl
• Figure 5: AFL