Contract-Based Incentive Mechanism for Federated Learning in Edge Computing System

In the edge computing system, federated learning (FL) is a distributed machine learning approach designed to allow multiple participants (e.g., mobile devices, edge nodes, or organizations) to collaborate on training machine learning models without sharing raw data. However, when training FL over wireless networks, mobile users (MU) need to transmit local model parameters over the wireless channel, which introduces training and transmission overheads and results in not enough MUs willing to participate in FL. In order to improve the performance of FL, it is necessary to introduce an appropriate incentive mechanism to encourage MUs to participate in the FL training task. In this paper, we adopt contract theory to design an effective incentive mechanism to motivate MUs to join FL. Edge computing base station (BS), maximize their own utility by signing contracts with MUs regarding contributions and rewards of the MU. We model the utility maximization of BS as a problem of solving the maximum value of a concave function and successfully find the optimal solution with the help of Lagrangian dyadic method. The simulation results show the proposed optimal contract satisfies individual rationality (IR) and incentive compatibility (IC), and it can also effectively improve the accuracy of FL, and obtain a high-quality global model with a fast speed.