网络与信息安全学报2025，Vol.11Issue(3)：19-44,26.DOI:10.11959/j.issn.2096-109x.2025035
激励驱动的联邦学习研究综述:隐私与安全
Survey on incentive-driven federated learning:privacy and security
摘要
Abstract
Federated learning was enabled to allow multiple data holders to jointly complete machine learning tasks without disclosing local data.Incentivizing participants to engage in federated learning and contribute high-quality data was identified as one of the key factors for its success.However,federated learning was found to face chal-lenges in both privacy leakage and security threats in practice.On the one hand,malicious participants were ob-served to launch active attacks to disrupt the process and results of federated learning,compromising its effective-ness and robustness.On the other hand,privacy leakage risks caused by passive attacks were shown to negatively impact participants'willingness to join federated learning,making the implementation of incentive mechanisms more difficult.In recent years,extensive research was conducted by the community on incentive-driven federated learning from the perspectives of privacy preservation and security defense,aiming to provide comprehensive solu-tions that balance security and fairness for federated learning.First,various passive and active attacks faced by incentive-driven federated learning were introduced,and the privacy leakage risks and security threats posed by these attacks were analyzed.Subsequently,a comprehensive review and analysis of incentive-driven federated learning research were provided from the perspectives of privacy preservation and security defense.In terms of pri-vacy preservation,the application of differential privacy and homomorphic encryption technologies in incentive-driven federated learning was focused on.In terms of security defense,various methods that combine incentive mechanisms to defend against poisoning attacks and free-riding attacks were specifically outlined.Finally,the chal-lenges still faced by incentive-driven federated learning in addressing privacy leakage and security attacks were dis-cussed,and potential future research directions in this field were explored.关键词
联邦学习/激励机制/隐私保护/差分隐私/博弈论Key words
federated learning/incentive mechanisms/privacy preservation/differential privacy/game theory分类
计算机与自动化引用本文复制引用
迟欢欢,熊平,刘恒竹,马霄,朱天清..激励驱动的联邦学习研究综述:隐私与安全[J].网络与信息安全学报,2025,11(3):19-44,26.基金项目
高等学校学科创新引智基地(B21038) （B21038）
湖北省自然科学基金(2025AFC108,2024AFB957) Innovation and Talent Base for Digital Technology and Finance(B21038) （2025AFC108,2024AFB957）
The Natural Science Foundation of Hubei Province(2025AFC108,2024AFB957) （2025AFC108,2024AFB957）
