Ensuring Fairness in Federated Learning Services: Innovative Approaches to Client Selection, Scheduling, and Rewards

Federated Learning (FL) Services enable customers (requesters) to outsource their FL tasks to the FL service provider, who will recruit a group of clients with appropriate datasets to complete the FL task. For a given FL task, how to select appropriate clients fairly becomes a challenging problem due to budget restrictions and client heterogeneity. In this paper, we propose a new client selection, scheduling, and rewarding scheme to ensure fairness through a three-stage process: 1) multicriteria initial client pool selection, 2) data quality-oriented perround client scheduling, and 3) performance-based rewarding. Specifically, we first define a client selection metric with multiple criteria, such as client resources, data quality, and client behaviors. Then, we formulate the initial client pool selection problem into an optimization problem that aims to maximize the overall scores of selected clients within a given budget and propose a greedy algorithm to solve it. Furthermore, we formulate the per-round client selection problem into a data quality-oriented scheduling problem that aims to improve model quality and guarantee fairness. We propose a heuristic algorithm to divide the pool into several subsets such that the federated dataset in a subset is close to an independent and identical distribution (iid) while guaranteeing each client is selected at least once in a scheduling period. In addition, we propose a performance-based payment adjustment protocol with a bonus and punishment mechanism, such that the final payment reflects the actual performance of each selected client. Our fairness analysis and experimental results show that our scheme not only can guarantee fairness but also can improve the model quality especially when data are non-iid.