[2309.13786] Distribution-Free Statistical Dispersion Management for Societal Purposes


Obtain a PDF of the paper titled Distribution-Free Statistical Dispersion Management for Societal Purposes, by Zhun Deng and 4 different authors

Obtain PDF
HTML (experimental)

Summary:Specific finite-sample statistical ensures on mannequin efficiency are an necessary ingredient in accountable machine studying. Earlier work has centered primarily on bounding both the anticipated lack of a predictor or the likelihood that a person prediction will incur a loss worth in a specified vary. Nevertheless, for a lot of high-stakes functions, it’s essential to grasp and management the dispersion of a loss distribution, or the extent to which completely different members of a inhabitants expertise unequal results of algorithmic choices. We provoke the research of distribution-free management of statistical dispersion measures with societal implications and suggest a easy but versatile framework that enables us to deal with a a lot richer class of statistical functionals past earlier work. Our strategies are verified by way of experiments in poisonous remark detection, medical imaging, and movie suggestion.

Submission historical past

From: Zhun Deng [view email]
Mon, 25 Sep 2023 00:31:55 UTC (10,494 KB)
Wed, 6 Mar 2024 14:18:52 UTC (10,186 KB)

Supply hyperlink


Please enter your comment!
Please enter your name here