The content presents a novel federated learning (FL) framework called QI-DPFL (Quality-Aware and Incentive-Boosted Federated Learning with Differential Privacy) that addresses the challenges of encouraging mobile edge devices to participate zealously in FL model training procedures while mitigating the privacy leakage risks during wireless transmission.
The key highlights of the framework are:
Client Selection Mechanism: A quality-aware client selection mechanism is proposed based on the Earth Mover's Distance (EMD) metric to select clients with high-quality datasets.
Incentive Mechanism Design: An incentive-boosted mechanism is designed that constructs the interactions between the central server and the selected clients as a two-stage Stackelberg game. The central server designs the time-dependent reward to minimize its cost by considering the trade-off between accuracy loss and total reward allocated, and each selected client decides the privacy budget to maximize its utility.
Differential Privacy Integration: The ρ-zero-concentrated differential privacy (ρ-zCDP) technique is integrated to obscure local model parameters and address privacy concerns during gradient propagation.
Stackelberg Nash Equilibrium Analysis: The optimal reward and the optimal privacy budget are derived for the central server and selected clients respectively, and it is proven that the optimal strategy profile forms a Stackelberg Nash Equilibrium.
Extensive experiments on different real-world datasets demonstrate the effectiveness of the proposed QI-DPFL framework in realizing the goal of privacy protection and incentive compatibility.
לשפה אחרת
מתוכן המקור
arxiv.org
תובנות מפתח מזוקקות מ:
by Wenhao Yuan,... ב- arxiv.org 04-15-2024
https://arxiv.org/pdf/2404.08261.pdfשאלות מעמיקות