基于标签的无数据的成员推理攻击  

Label-based data-free membership inference attack

作　　者：杨盼盼 张信明[1] Yang Panpan;Zhang Xinming(School of Computer Science and Technology,University of Science and Technology of China,Hefei 230026,China)

机构地区：[1]中国科学技术大学计算机科学与技术学院,安徽合肥230026

出　　处：《网络安全与数据治理》2023年第5期44-49,共6页CYBER SECURITY AND DATA GOVERNANCE

基　　金：国家重点研究发展计划(2020YFB2103803)。

摘　　要：成员推理攻击根据模型的预测结果推断特定记录是否为模型的训练数据成员,在隐私保护等领域具有重要应用意义。现有的基于标签的无数据的成员推理攻击方法主要利用对抗样本技术,存在查询和计算成本较高的问题。对此提出一种新的成员推理攻击方法,该方法利用影子模型来减少多次攻击的查询代价,并提出数据筛选与优化策略以提高攻击模型的性能。实验在两个常见的图像数据集上进行,结果表明该方法同时具有较高的攻击成功率和较低的查询成本。Membership inference attack infers whether a specific record is a training data member of a model based on the model′s prediction results,which has important application significance in the field of privacy protection.Existing label-based data-free membership inference attack methods mainly use adversarial sample technology,which has a high query and computation cost.This paper proposed a new membership inference attack method that uses a shadow model to reduce the query cost of multiple attacks,and proposed data filtering and optimization strategies to improve the performance of the attack model.Experiments were conducted on two commonly used image datasets,and the results showed that the proposed method has both a high attack success rate and a low query cost.

关 键 词：成员推理攻击 无数据 基于标签 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]