PI-FL: Personalized and Incentivized Federated Learning

Ahmad Faraz Khan, Xinran Wang, Qi Le, Azal Ahmad Khan, Haider Ali, Jie Ding, Ali Anwar, Ali R. Butt

Keywords: Federated Learning, Personalized Federated Learning, Incentivized Federated Learning, Client Autonomy

TL;DR: The paper proposes an incentive mechanism that fosters personalized Federated Learning and enhances the personalized and cluster-level model appeal for clients, leading to active and consistent participation and improvement in test accuracy.

Abstract: Existing incentive solutions for traditional Federated Learning (FL) only consider individual clients' contributions to a single global model. They are unsuitable for clustered personalization, where multiple cluster-level models can exist. Moreover, they focus solely on providing monetary incentives and fail to address the need for personalized FL, overlooking the importance of enhancing the personalized model's appeal to individual clients as a motivating factor for consistent participation. In this paper, we first propose to treat incentivization and personalization as interrelated challenges and solve them with an incentive mechanism that fosters personalized learning. Second, unlike existing approaches that rely on the aggregator to perform client clustering, we propose to involve clients by allowing them to provide incentive-driven preferences for joining clusters based on their data distributions. Our approach enhances the personalized and cluster-level model appeal for self-aware clients with high-quality data leading to their active and consistent participation. Through evaluation, we show that we achieve an 8-45% test accuracy improvement of the cluster models, 3-38% improvement in personalized model appeal, and 31-100% increase in the participation rate, compared to a wide range of FL modeling approaches, including those that tackle data heterogeneity and learn personalized models.

Submission Number: 8896