[Paper Review] Communication Efficient Federated Learning with Energy Awareness over Wireless Networks

Richeng Jin, Xiaofan He+1 more|arXiv (Cornell University)|Apr 15, 2020

Privacy-Preserving Technologies in Data40 references4 citations

TL;DR

This paper proposes an energy-aware, communication-efficient federated learning framework for wireless networks that uses SignSGD to transmit only gradient signs, reducing bandwidth usage. By optimizing transmission power and rate under partial CSI (only at the server), it balances learning performance and device energy consumption, achieving improved convergence under realistic wireless outage conditions and data heterogeneity.

ABSTRACT

In federated learning (FL), reducing the communication overhead is one of the most critical challenges since the parameter server and the mobile devices share the training parameters over wireless links. With such consideration, we adopt the idea of SignSGD in which only the signs of the gradients are exchanged. Moreover, most of the existing works assume Channel State Information (CSI) available at both the mobile devices and the parameter server, and thus the mobile devices can adopt fixed transmission rates dictated by the channel capacity. In this work, only the parameter server side CSI is assumed, and channel capacity with outage is considered. In this case, an essential problem for the mobile devices is to select appropriate local processing and communication parameters (including the transmission rates) to achieve a desired balance between the overall learning performance and their energy consumption. Two optimization problems are formulated and solved, which optimize the learning performance given the energy consumption requirement, and vice versa. Furthermore, considering that the data may be distributed across the mobile devices in a highly uneven fashion in FL, a stochastic sign-based algorithm is proposed. Extensive simulations are performed to demonstrate the effectiveness of the proposed methods.

Motivation & Objective

• Address the challenge of high communication overhead and energy consumption in federated learning over wireless networks.
• Overcome the limitation of existing works that assume full channel state information (CSI) at both clients and server.
• Optimize the trade-off between learning performance and energy consumption under wireless channel outages and data heterogeneity.
• Develop a stochastic sign-based algorithm to handle non-iid data distributions across mobile devices.
• Formulate and solve two optimization problems: maximizing learning performance under energy constraints, and minimizing energy use under performance requirements.

Proposed method

• Adopt SignSGD with majority vote to transmit only the signs of model gradients, drastically reducing communication overhead.
• Model wireless uplink transmission using flat-fading channels with outage probability, assuming CSI is only available at the parameter server.
• Formulate two optimization problems: (1) maximize convergence rate under a given energy budget, and (2) minimize energy consumption under a required convergence rate.
• Derive closed-form expressions for optimal transmission power and rate per device based on local computation cost, data size, and channel conditions.
• Introduce a stochastic sign-based algorithm to handle data heterogeneity, where clients update gradients based on local data distributions.
• Propose a new performance metric combining per-round aggregation success probability and convergence rate to evaluate practical learning performance.

Experimental results

Research questions

• RQ1 How does wireless channel outage affect the aggregation accuracy of sign-based gradient updates in federated learning?
• RQ2 What is the optimal trade-off between communication energy and local computation energy under wireless constraints?
• RQ3 How can transmission power and rate be jointly optimized at mobile devices when only the server has CSI?
• RQ4 How does data heterogeneity impact the convergence of sign-based federated learning under wireless constraints?
• RQ5 Can a new performance metric that combines outage probability and convergence rate better reflect real-world training efficiency?

Key findings

• The proposed method achieves a significant reduction in communication overhead by transmitting only gradient signs, enabling efficient training over bandwidth-constrained wireless links.
• Optimal transmission power and rate are derived in closed form, balancing energy consumption and learning performance under partial CSI.
• The stochastic sign-based algorithm effectively handles data heterogeneity, maintaining stable convergence even with non-iid data distributions.
• The proposed performance metric—combining per-round aggregation success and convergence rate—provides a more accurate measure of practical learning efficiency than round count alone.
• Simulations show that the optimized system achieves faster convergence and lower energy consumption compared to baseline methods under realistic wireless outage conditions.
• The analysis proves that the probability of correct sign aggregation increases with the number of clients and decreases with individual outage probability, validating the robustness of the approach.