福建电脑2025，Vol.41Issue(3)：20-23,4.DOI:10.16707/j.cnki.fjpc.2025.03.005
抗中毒攻击的鲁棒隐私保护模型
The Robust Privacy Protection Model Against Poisoning Attacks
李楠 1叶嘉宾 2蒋彦甫 3孙祥 4刘雪慧4
作者信息
- 1. 南京大数据集团有限公司 江苏 南京 210000||东南大学 江苏 南京 210000
- 2. 南京大数据集团有限公司 江苏 南京 210000||南京智能计算科技发展有限公司 江苏 南京 210000
- 3. 集美大学 福建 厦门 361021
- 4. 南京智能计算科技发展有限公司 江苏 南京 210000
- 折叠
摘要
Abstract
Federated learning is susceptible to malicious gradient poisoning attacks in terms of privacy protection,and existing defense strategies suffer from high computational and communication costs.To address these issues,this article proposes a robust privacy protection model against viral attacks.This model evaluates encryption gradients through internal audit mechanisms,uses Gaussian mixture models combined with Mahalanobis distance for robust aggregation,and employs additive homomorphic encryption technology to ensure data security.The experimental results show that the model proposed in this paper can effectively resist attacks and significantly reduce the cost of computation and communication,demonstrating better privacy and accuracy protection compared to other encryption techniques.关键词
隐私保护/中毒攻击/鲁棒隐私保护模型Key words
Privacy Protection/Poisoning Attack/Robust Privacy Protection Model分类
林学引用本文复制引用
李楠,叶嘉宾,蒋彦甫,孙祥,刘雪慧..抗中毒攻击的鲁棒隐私保护模型[J].福建电脑,2025,41(3):20-23,4.
