FedTIME: a two-tier multi-contract federated learning incentive mechanism with time cost

Résumé fourni par la source

Federated Learning faces challenges such as insufficient client participation, motivation, and inefficient resource allocation, which, along with issues like device heterogeneity, constrain the model training effectiveness. This paper proposes a federated learning incentive mechanism that integrates time cost, dynamic reputation evaluation, and hierarchical contract design (Time-sensitive Contract-based Incentive Mechanism, TCIM), and constructs the federated learning framework FedTIME. This mechanism is modelled through a three-dimensional cost function, combined with Stackelberg game theory for dynamic reward allocation optimization, ultimately effectively suppressing malicious behavior, enhancing model convergence efficiency, and social welfare. Theoretical analysis proves that the mechanism satisfies six properties including incentive compatibility and privacy preservation. Experimental evaluation on benchmark datasets demonstrates the framework's superior performance in mitigating client procrastination and improving overall model performance and system welfare.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.