Aging-aware Incentive Mechanism Design for Federated Learning in Industrial Internet of Things

Abstract

The privacy concern of the sensory data at Industrial Internet-of-Things (IIoT) devices hinders the rapid development of intelligent IIoT. Federated learning (FL) is widely adopted to preserve data privacy. However, the lack of participants jeopardizes the practicality of FL. This paper designs an efficient incentive mechanism for FL to encourage more IIoT devices to join the FL tasks. Considering the real-time demand of data processing for some IIoT applications, we first propose an aging-aware reputation evaluation scheme based on Age of Information to select high-quality devices. We then propose an efficient incentive mechanism based on contract theory with asymmetric information. The proposed approach enables the edge server to provide appropriate rewards by offering a tailored contract to each IIoT device based on their contribution. Simulation results demonstrate that with the increase of reputation values and the size of data samples, the utilities of the edge server and devices can be greatly improved respectively.