Matching Game for Multi-Task Federated Learning in Internet of Vehicles

To overcome the inherent defects of massive data uploading and processing in traditional machine learning, federated learning is emerged as a promising tool given that it enables to implement privacy-preserved distributed machine learning in Internet of Vehicles (IoV). However, the performance of federated learning suffers from several challenges, especially ineffective execution of delay-sensitive tasks triggered simultaneously by moving vehicles. To minimize the total execution delay of multiple tasks, we propose a multi-task federated learning framework which improves task completion rate and enables each task to be completed in time. Moreover, we also aim to improve the network utility of the IoV. The algorithm of joint optimization algorithm is proposed to achieve a stable partition of vehicle coalitions based on the block coordinate descent (BCD) method, the matching game-based coalition method, and gradient projection method. The performance of the proposed multi-task federated learning is evaluated through numerical simulations in terms of total latency, network utility, and accuracy of federated learning tasks. The results show that our proposed multi-task federated learning framework and algorithm guarantees the completion of multiple delay-sensitive tasks effectively while improving vehicular network utility.