Abstract:Federated Learning (FL) has emerged as a solution for distributed model training across decentralized, privacy-preserving devices, but the different energy capacities of participating devices (system heterogeneity) constrain real-world implementations. These energy limitations not only reduce model accuracy but also increase dropout rates, impacting on convergence in practical FL deployments. In this work, we propose LeanFed, an energy-aware FL framework designed to optimize client selection and training workloads on battery-constrained devices. LeanFed leverages adaptive data usage by dynamically adjusting the fraction of local data each device utilizes during training, thereby maximizing device participation across communication rounds while ensuring they do not run out of battery during the process. We rigorously evaluate LeanFed against traditional FedAvg on CIFAR-10 and CIFAR-100 datasets, simulating various levels of data heterogeneity and device participation rates. Results show that LeanFed consistently enhances model accuracy and stability, particularly in settings with high data heterogeneity and limited battery life, by mitigating client dropout and extending device availability. This approach demonstrates the potential of energy-efficient, privacy-preserving FL in real-world, large-scale applications, setting a foundation for robust and sustainable pervasive AI on resource-constrained networks.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the model performance degradation and device dropout issues in Federated Learning (FL) due to the energy limitations of participating devices. Specifically: 1. **Energy Heterogeneity**: Different devices have different energy capacities, which restricts the effectiveness of distributed model training in actual deployments. Energy limitations not only reduce the accuracy of the model but also increase the device dropout rate, affecting the convergence of federated learning. 2. **Device Dropout and Gradient Divergence**: A client - selection strategy that ignores battery life may accelerate battery depletion, causing devices to be temporarily or permanently unavailable during training, resulting in a significant performance degradation. 3. **Data Heterogeneity**: Data in the real world is usually non - independently and identically distributed (non - iid) among clients, and only some devices participate in each training round. Therefore, the convergence depends on the number and specific selection of participating devices. To solve these problems, the authors propose an energy - aware federated learning framework named LeanFed, aiming to optimize client selection and training workload, especially for applications on battery - constrained devices. By dynamically adjusting the proportion of local data used by each device during the training process, LeanFed ensures that devices do not run out of power during the process, thereby maximizing device participation and improving the accuracy and stability of the model. ### Specific Measures: - **Adaptive Data Usage**: Dynamically adjust the proportion of local data used by each device during training to extend the battery life of the device and ensure that it remains active throughout the communication round. - **Reduce Device Dropout**: By controlling the amount of data used by each device, reduce device dropout due to battery depletion, thereby ensuring more stable model convergence. - **Improve Model Performance**: Experimental results show that LeanFed can significantly improve the accuracy and stability of the model in the case of high data heterogeneity and limited battery life. In conclusion, the core problem of this paper is to solve the performance challenges in federated learning due to device energy limitations and proposes a new framework to address these challenges.

Learn More by Using Less: Distributed Learning with Energy-Constrained Devices

Towards Energy-Aware Federated Learning on Battery-Powered Clients

AEDFL: Efficient Asynchronous Decentralized Federated Learning with Heterogeneous Devices

Efficient and Less Centralized Federated Learning

Enhancing Efficiency in Multidevice Federated Learning through Data Selection

Adaptive Decentralized Federated Learning in Energy and Latency Constrained Wireless Networks

FedLE: Federated Learning Client Selection with Lifespan Extension for Edge IoT Networks

AnycostFL: Efficient On-Demand Federated Learning over Heterogeneous Edge Devices

An Energy and Carbon Footprint Analysis of Distributed and Federated Learning

D2D-Assisted Adaptive Federated Learning in Energy-Constrained Edge Computing

Energy-Aware Heterogeneous Federated Learning via Approximate DNN Accelerators

Efficient Data Distribution Estimation for Accelerated Federated Learning

CEFL: Carbon-Efficient Federated Learning

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

Enhancing Federated Learning Convergence with Dynamic Data Queue and Data Entropy-driven Participant Selection

Energy-Aware Edge Association for Cluster-based Personalized Federated Learning

Federated Learning with Pareto Optimality for Resource Efficiency and Fast Model Convergence in Mobile Environments

SCALE: Self-regulated Clustered federAted LEarning in a Homogeneous Environment

Flexible Clustered Federated Learning for Client-Level Data Distribution Shift

FedAda: Fast-convergent adaptive federated learning in heterogeneous mobile edge computing environment

Energy Efficient Federated Learning over Heterogeneous Mobile Devices via Joint Design of Weight Quantization and Wireless Transmission