arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2305.03355cs.LGcs.AI
收藏

关于数据集蒸馏的综合研究:性能、隐私、鲁棒性与公平性

A Comprehensive Study on Dataset Distillation: Performance, Privacy, Robustness and Fairness

Fraunhofer FOKUS(弗劳恩霍夫开放通信系统研究所);University of Stavanger(斯塔万格大学);Technische Universität München(慕尼黑工业大学)

Zongxiong Chen, Jiahui Geng, Derui Zhu, Herbert Woisetschlaeger, Qing Li, Sonja Schimmler, Ruben Mayer, Chunming Rong

AI总结:

本文从安全视角系统评估数据集蒸馏方法,发现其仍存在隐私泄露风险,并会削弱模型鲁棒性、加剧类别间不公平性,同时提出大规模评测基准框架。

摘要:

数据集蒸馏的目标是将原始数据集的丰富特征编码到一个微型数据集中。这是一种加速神经网络训练及相关研究的有前景的方法。已有多种方法被提出以提升蒸馏图像的信息量和泛化性能。然而,目前尚无工作从安全角度对该技术进行全面分析,也缺乏对其潜在风险的系统性理解。在本工作中,我们进行了大量实验以评估当前最先进的数据集蒸馏方法。我们成功利用成员推断攻击表明隐私风险仍然存在。我们的工作还证明,数据集蒸馏会对模型鲁棒性造成不同程度的影响,并在进行预测时放大模型在不同类别间的公平性问题。本工作为数据集蒸馏评估提供了一个大规模基准测试框架。

英文摘要:

The aim of dataset distillation is to encode the rich features of an original dataset into a tiny dataset. It is a promising approach to accelerate neural network training and related studies. Different approaches have been proposed to improve the informativeness and generalization performance of distilled images. However, no work has comprehensively analyzed this technique from a security perspective and there is a lack of systematic understanding of potential risks. In this work, we conduct extensive experiments to evaluate current state-of-the-art dataset distillation methods. We successfully use membership inference attacks to show that privacy risks still remain. Our work also demonstrates that dataset distillation can cause varying degrees of impact on model robustness and amplify model unfairness across classes when making predictions. This work offers a large-scale benchmarking framework for dataset distillation evaluation.

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。

↑