【Data Privacy顶会论文笔记汇总】
发布时间
阅读量：
阅读量
Data Privacy顶会论文笔记汇总
联邦学习调研benchmark 汇总
一. 属性推理攻击
- 联邦学习或协作学习 + 执行主任务(target attribute)的同时 推测参与者的训练数据是否有目标属性
1.Exploiting Unintended Feature Leakage in Collaborative Learning [SP19]
2.Honest-but-Curious Nets Sensitive Attributes of Private Inputs [CCS2021]
- 纵向联邦学习 + 对被动方发起特征推理攻击
3.Feature inference attack on model predictions in vertical federated learning [ICDE2021]
- 纵向联邦学习 + 对主动方发起标签推理攻击
4.Label Inference Attacks Against Vertical Federated Learning. [usenix22]
二. 成员推理攻击
- 机器学习 ML + 推断成员样本是否在目标模型(ML)的训练数据集中
5.Membership Inference Attacks Against Machine Learning Models [SP17]
- 生成对抗网络GAN + 推断成员样本是否在目标模型GAN的训练数据集中
6.GAN-Leaks: A Taxonomy of Membership Inference Attacks against Generative Models [CCS20]
三. 梯度反转攻击
- 联邦学习 + 梯度反转重构参与方训练数据
7.Inverting Gradients - How easy is it to break privacy in federated learning? [CVPR22]
全部评论 (0)
还没有任何评论哟~