Journal Article
Machine Learning

Energy Efficient Federated Learning Over Wireless Communication Networks

Zhaohui Yang(King's College London), Mingzhe Chen(Princeton University), Walid Saad(Virginia Tech), Choong Seon Hong(Kyung Hee University), Mohammad Shikh‐Bahaei(King's College London)
November 20, 2020IEEE Transactions on Wireless Communications1,109 citations
arXivDOI

1.1k

Citations

85

Influential Citations

IEEE Transactions on Wireless Communications

Venue

2020

Year

Abstract

In this paper, the problem of energy efficient transmission and computation resource allocation for federated learning (FL) over wireless communication networks is investigated. In the considered model, each user exploits limited local computational resources to train a local FL model with its collected data and, then, sends the trained FL model to a base station (BS) which aggregates the local FL model and broadcasts it back to all of the users. Since FL involves an exchange of a learning model between users and the BS, both computation and communication latencies are determined by the learning accuracy level. Meanwhile, due to the limited energy budget of the wireless users, both local computation energy and transmission energy must be considered during the FL process. This joint learning and communication problem is formulated as an optimization problem whose goal is to minimize the total energy consumption of the system under a latency constraint. To solve this problem, an iterative algorithm is proposed where, at every step, closed-form solutions for time allocation, bandwidth allocation, power control, computation frequency, and learning accuracy are derived. Since the iterative algorithm requires an initial feasible solution, we construct the completion time minimization problem and a bisection-based algorithm is proposed to obtain the optimal solution, which is a feasible solution to the original energy minimization problem. Numerical results show that the proposed algorithms can reduce up to 59.5% energy consumption compared to the conventional FL method.

Analysis

Why This Paper Matters

This paper addresses a critical bottleneck in deploying federated learning (FL) on wireless edge devices: energy consumption. As FL relies on iterative local training and model transmission, both computation and communication drain limited battery resources. The authors systematically model and optimize these competing costs under a latency constraint, offering a principled approach that is directly applicable to real-world IoT and mobile networks. The 59.5% energy reduction over conventional methods underscores the practical significance of joint resource allocation.

By deriving closed-form solutions for key parameters—time, bandwidth, power, frequency, and accuracy—the paper moves beyond heuristic tuning. This analytical tractability makes the framework scalable and interpretable, which is valuable for system designers. The work also bridges the gap between communication theory and machine learning, a growing area of importance as AI moves to the edge.

Technical Contributions

  • Joint optimization formulation: Integrates computation energy (CPU cycles) and transmission energy (power, bandwidth) with learning accuracy as a decision variable, under a total latency constraint.
  • Iterative algorithm with closed-form updates: Each iteration solves for one variable while fixing others, yielding explicit expressions that avoid numerical solvers.
  • Bisection-based initialization: Solves the completion time minimization problem to provide a feasible starting point for the energy minimization, ensuring convergence.
  • Explicit modeling of learning accuracy: Accuracy is treated as a continuous variable linked to local iterations and data size, enabling trade-off analysis between model quality and energy.

Results

  • Up to 59.5% reduction in total energy consumption compared to a baseline FL method that uses fixed resource allocation.
  • The proposed algorithm consistently outperforms baselines across varying numbers of users, data sizes, and latency constraints.
  • The bisection-based initialization ensures feasibility and accelerates convergence of the iterative energy minimization.

Significance

This paper is among the first to rigorously formulate and solve the energy-latency trade-off in wireless FL. Its closed-form solutions and iterative framework have influenced subsequent work on resource allocation for distributed learning, including heterogeneous and multi-cell extensions. For practitioners, the results provide clear guidelines: adaptively tuning computation frequency, transmission power, and bandwidth can yield substantial energy savings without sacrificing model accuracy. The work also highlights the importance of co-designing learning algorithms with communication protocols, a theme that continues to drive research in edge AI.

federated-learningenergy-efficiencywireless-networksresource-allocationoptimizationlatency-constraintcomputation-communication

Related Papers