arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2201.04018cs.CRcs.AI
收藏

针对差分私有分裂学习的特征空间劫持攻击

Feature Space Hijacking Attacks against Differentially Private Split Learning

LiveRamp

Grzegorz Gawron, Philip Stubbings

摘要:

拆分学习和差分隐私技术在帮助分布式数据集进行隐私合规高级分析方面具有越来越大的潜力。对分裂学习的攻击是一种重要的评估工具,最近受到越来越多的研究关注。这项工作的贡献是应用最近的特征空间劫持攻击(FSHA)的分裂神经网络的学习过程中增强了差分隐私(DP),使用客户端现成的DP优化器。FSHA攻击在任意设置的DP攻击级别下以低出错率获得客户端的私有数据重建。我们还尝试了降维作为一种潜在的攻击风险缓解,并表明它可能在一定程度上有所帮助。我们讨论了为什么差分隐私在这种情况下不是一种有效的保护,并提到潜在的其他风险缓解方法的原因。

英文摘要:

Split learning and differential privacy are technologies with growing potential to help with privacy-compliant advanced analytics on distributed datasets. Attacks against split learning are an important evaluation tool and have been receiving increased research attention recently. This work's contribution is applying a recent feature space hijacking attack (FSHA) to the learning process of a split neural network enhanced with differential privacy (DP), using a client-side off-the-shelf DP optimizer. The FSHA attack obtains client's private data reconstruction with low error rates at arbitrarily set DP epsilon levels. We also experiment with dimensionality reduction as a potential attack risk mitigation and show that it might help to some extent. We discuss the reasons why differential privacy is not an effective protection in this setting and mention potential other risk mitigation methods.

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。

↑