AISTATS 2024poster2 citations
To Pool or Not To Pool: Analyzing the Regularizing Effects of Group-Fair Training on Shared Models
Cyrus Cousins, I. Elizabeth Kumar, Suresh Venkatasubramanian
Abstract
In fair machine learning, one source of performance disparities between groups is overfitting to groups with relatively few training samples. We derive group-specific bounds on the generalization error of welfare-centric fair machine learning that benefit from the larger sample size of the majority group. We do this by considering group-specific Rademacher averages over a restricted hypothesis class, which contains the family of models likely to perform well with respect to a fair learning objective (e.g., a power-mean). Our simulations demonstrate these bounds improve over a naïve method, as expected by theory, with particularly significant improvement for smaller group sizes.
BibTeX
@InProceedings{pmlr-v238-cousins24a,
title = {To Pool or Not To Pool: Analyzing the Regularizing Effects of Group-Fair Training on Shared Models},
author = {Cousins, Cyrus and Elizabeth Kumar, I. and Venkatasubramanian, Suresh},
booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics},
pages = {4573--4581},
year = {2024},
editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen},
volume = {238},
series = {Proceedings of Machine Learning Research},
month = {02--04 May},
publisher = {PMLR},
pdf = {https://proceedings.mlr.press/v238/cousins24a/cousins24a.pdf},
url = {https://proceedings.mlr.press/v238/cousins24a.html},
abstract = {In fair machine learning, one source of performance disparities between groups is overfitting to groups with relatively few training samples. We derive group-specific bounds on the generalization error of welfare-centric fair machine learning that benefit from the larger sample size of the majority group. We do this by considering group-specific Rademacher averages over a restricted hypothesis class, which contains the family of models likely to perform well with respect to a fair learning objective (e.g., a power-mean). Our simulations demonstrate these bounds improve over a naïve method, as expected by theory, with particularly significant improvement for smaller group sizes.}
}