An Incentive Mechanism for Federated Learning in IIoT: A Long-Term Perspective
Chu Wang, Shuo He, Shupan Li, Longlong Xing, Saisai Zhu, Hongen Xu · 2024
Federated learning (FL) is a new machine learning paradigm that can protect data privacy while enabling data-driven insights. Its performance heavily relies on the quality of local models that participants contribute. However, devices may be reluctant to contribute their resources without sufficient incentives. A carefully crafted incentive mechanism is crucial to the practical application of FL. However, existing studies overlook the long-term performance, the requirement of data freshness in some Industrial Internet of Things (IIoT) applications, and the capability that IIoT devices can engage in multiple tasks simultaneously. To address these issues, we formulate the incentive mechanism as a stochastic optimization problem which minimizes the long-term energy cost of IIoT devices. Lyapunov optimization is employed to design an online winner selection and resource allocation algorithm. Moreover, we consider the data freshness as a criterion to design reward for devices to motivate their participation. Simulation results show that the proposed mechanism can effectively reduce the average energy cost of devices and ensure continuous participation.