arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2306.05068cs.LGcs.AIcs.CY
收藏

揭示机器学习中的代表性不足与抽样偏差

Shedding light on underrepresentation and Sampling Bias in machine learning

INRIA(法国国家信息与自动化研究所);École Polytechnique(巴黎综合理工学院)

Sami Zhioua, Rūta Binkytė

AI总结:

本文辨析机器学习中的抽样偏差,提出样本量偏差(SSB)与代表性不足偏差(URB),并将歧视分解为方差、偏差和噪声,质疑单纯增加少数群体样本即可缓解歧视的观点。

摘要:

准确衡量歧视对于忠实地评估训练后的机器学习(ML)模型的公平性至关重要。衡量歧视时存在的任何偏差,都会放大或低估既有的差异。偏差有若干来源,而且通常假定机器学习产生的偏差由不同群体(例如女性与男性、白人与黑人等)平等承担。然而,如果偏差由不同群体以不同方式承担,它可能加剧对特定子群体的歧视。在文献中,sampling bias(抽样偏差)这一术语被不一致地用于描述由抽样过程导致的偏差。本文中,我们尝试通过引入定义明确的抽样偏差变体来消除该术语的歧义,即 sample size bias(SSB,样本量偏差)和 underrepresentation bias(URB,代表性不足偏差)。我们还展示了如何将歧视分解为方差、偏差和噪声。最后,我们质疑了一种被广泛接受的缓解方法,即认为可以通过收集更多代表性不足群体的样本,来解决歧视问题。

英文摘要:

Accurately measuring discrimination is crucial to faithfully assessing fairness of trained machine learning (ML) models. Any bias in measuring discrimination leads to either amplification or underestimation of the existing disparity. Several sources of bias exist and it is assumed that bias resulting from machine learning is born equally by different groups (e.g. females vs males, whites vs blacks, etc.). If, however, bias is born differently by different groups, it may exacerbate discrimination against specific sub-populations. Sampling bias, is inconsistently used in the literature to describe bias due to the sampling procedure. In this paper, we attempt to disambiguate this term by introducing clearly defined variants of sampling bias, namely, sample size bias (SSB) and underrepresentation bias (URB). We show also how discrimination can be decomposed into variance, bias, and noise. Finally, we challenge the commonly accepted mitigation approach that discrimination can be addressed by collecting more samples of the underrepresented group.

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。

↑