Incentive design improves fairness and efficiency in wireless federated learning
DisC: Incentive Design for Green, Robust, and Fair Client Self-Participation in OTA Federated Learning
Networking and Internet ArchitectureComputer Science and Game Theory
Summary
Wireless federated learning allows many devices to train a shared model together by sending updates over the air, but different device conditions can make this process unfair and less accurate. The authors propose DisC, a system where each device decides how much to participate based on its own channel quality, energy, and contribution to the learning. This approach encourages balanced, efficient participation and leads to stable outcomes where the learning process runs smoothly. DisC works without a central controller, making it practical for real-world wireless networks.
What this means in practice
- •For wireless network engineers: Manage device participation in over-the-air federated learning to balance channel conditions and energy for fairer model updates.
- •For mobile device manufacturers: Incorporate distributed incentive mechanisms to prolong battery life while contributing to federated learning models via wireless channels.
Authors
Driss Choukri, Essaid Sabir, Abdelkrim Haqiq
Abstract
Over-the-Air Federated Learning (OTA-FL) improves communication efficiency by exploiting waveform superposition over wireless Multiple-Access Channels (MAC). However, client heterogeneity, fading, energy constraints, and participation imbalance can degrade both fairness and aggregation accuracy. Existing Pareto-oriented fair OTA-FL schemes may still suffer from large aggregation distortion when highly weighted clients experience poor channel conditions. This paper proposes DisC, a game-theoretic distributed client selection framework for fairness-aware, channel-aware, and energy-efficient OTA-FL. Each client selects its participation level by maximizing a utility that captures Pareto fairness importance, learning relevance, OTA distortion, energy consumption, and delay. We show that DisC admits a Nash equilibrium and can be formulated as an exact potential game, where stable equilibria correspond to locally optimal OTA-FL operating points. DisC therefore offers a distributed, incentive-compatible alternative to centralized client selection under wireless and resource constraints.