arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2404.15760cs.LGcs.AIstat.ML
收藏

使用反事实样本消除机器遗忘的偏差

Debiasing Machine Unlearning with Counterfactual Examples

Stony Brook University(石溪大学);Walmart Global Tech(沃尔玛全球科技);The Xi’an Jiaotong-Liverpool University(西交利物浦大学);Boise State University(博伊西州立大学);Indian Institute of Technology Hyderabad(印度理工学院海得拉巴分校);Sapienza University of Rome(罗马大学);Zhejiang University(浙江大学);Institute of semiconductors, Chinese Academy of Sciences(中国科学院半导体研究所)

Ziheng Chen, Jia Wang, Jun Zhuang, Abbavaram Gowtham Reddy, Fabrizio Silvestri, Jin Huang, Kaushiki Nag, Kun Kuang, Xin Ning, Gabriele Tolomei

AI总结:

本研究分析了机器遗忘过程中的数据级与算法级偏差成因,提出基于干预的去偏数据集擦除方法,并利用反事实样本指导遗忘过程以保持语义一致性,在评估指标上优于现有基线。

摘要:

被遗忘权(RTBF)旨在通过实施机器学习技术来保护个人免受其历史行为的持久影响。这些技术能够在无需大规模模型重训练的情况下,删除已获取的知识。然而,它们往往忽略了一个关键问题:遗忘过程偏差。这种偏差源于两个主要方面:(1)数据级偏差,表现为数据删除不均匀;(2)算法级偏差,导致剩余数据集被污染,从而降低模型准确率。在本研究中,我们分析了遗忘过程背后的因果因素,并在数据和算法两个层面缓解了偏差。具体而言,我们引入了一种基于干预的方法,使用去偏数据集来擦除需要遗忘的知识。此外,我们利用反事实样本来指导遗忘过程,因为它们能在不损害剩余数据集性能的情况下保持语义数据一致性。实验结果表明,我们的方法在评估指标上优于现有的机器遗忘基线方法。

英文摘要:

The right to be forgotten (RTBF) seeks to safeguard individuals from the enduring effects of their historical actions by implementing machine-learning techniques. These techniques facilitate the deletion of previously acquired knowledge without requiring extensive model retraining. However, they often overlook a critical issue: unlearning processes bias. This bias emerges from two main sources: (1) data-level bias, characterized by uneven data removal, and (2) algorithm-level bias, which leads to the contamination of the remaining dataset, thereby degrading model accuracy. In this work, we analyze the causal factors behind the unlearning process and mitigate biases at both data and algorithmic levels. Typically, we introduce an intervention-based approach, where knowledge to forget is erased with a debiased dataset. Besides, we guide the forgetting procedure by leveraging counterfactual examples, as they maintain semantic data consistency without hurting performance on the remaining dataset. Experimental results demonstrate that our method outperforms existing machine unlearning baselines on evaluation metrics.

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。

↑