arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2212.07591cs.LGcs.AIcs.CR
收藏

剖析分布推断

Dissecting Distribution Inference

University of Virginia(弗吉尼亚大学);University of Michigan(密歇根大学)

Anshuman Suri, Yifu Lu, Yanjin Chen, David Evans

AI总结:

本文提出一种在多数场景下优于已知最佳白盒攻击的新型黑盒分布推断攻击,并据此评估不同攻击者知识假设及防御,发现简单重采样防御高度有效。

摘要:

分布推断攻击旨在推断用于训练机器学习模型的数据的统计属性。这类攻击有时具有惊人的强大效力,但影响分布推断风险的因素尚未得到充分理解,而且已展示的攻击往往依赖强烈且不现实的假设,例如即使在所谓的黑盒威胁场景中也完全掌握训练环境。为了增进对分布推断风险的理解,我们开发了一种新的黑盒攻击,在大多数设置下甚至优于已知最佳的白盒攻击。利用这种新攻击,我们在放宽关于攻击者在黑盒访问下所掌握知识的多种假设(例如已知模型架构和仅标签访问)的同时,评估了分布推断风险。最后,我们评估了先前提出的防御措施的有效性,并引入了新的防御措施。我们发现,尽管基于噪声的防御似乎无效,但一种简单的重采样防御可以非常有效。代码可在 https://github.com/iamgroot42/dissecting_distribution_inference 获取。

英文摘要:

A distribution inference attack aims to infer statistical properties of data used to train machine learning models. These attacks are sometimes surprisingly potent, but the factors that impact distribution inference risk are not well understood and demonstrated attacks often rely on strong and unrealistic assumptions such as full knowledge of training environments even in supposedly black-box threat scenarios. To improve understanding of distribution inference risks, we develop a new black-box attack that even outperforms the best known white-box attack in most settings. Using this new attack, we evaluate distribution inference risk while relaxing a variety of assumptions about the adversary's knowledge under black-box access, like known model architectures and label-only access. Finally, we evaluate the effectiveness of previously proposed defenses and introduce new defenses. We find that although noise-based defenses appear to be ineffective, a simple re-sampling defense can be highly effective. Code is available at https://github.com/iamgroot42/dissecting_distribution_inference

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑