BriefGPT.xyz
Feb, 2025
通过最大子群差异检测偏见
Bias Detection via Maximum Subgroup Discrepancy
HTML
PDF
Jiří Němeček, Mark Kozdoba, Illia Kryvoviaz, Tomáš Pevný, Jakub Mareček
TL;DR
本文解决了经典度量方法(如总变差和Wasserstein距离)在偏见评估中样本复杂度高的问题。我们提出了一种新颖的度量方法——最大子群差异(MSD),能够以线性样本复杂度评估特征子群的偏差,并提供实用的算法和清晰的偏差修复路径。实证结果表明,MSD在真实数据集上的表现优于现有其他度量方法。
Abstract
Bias evaluation is fundamental to trustworthy AI, both in terms of checking
Data Quality
and in terms of checking the outputs of AI systems. In testing
Data Quality
, for example, one may study a distance of a giv
→