To Pool or Not to Pool: Analyzing the Regularizing Effects of Group-Fair Training on Shared Models

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.

Cite

Text

Cousins et al. "To Pool or Not to Pool: Analyzing the Regularizing Effects of Group-Fair Training on Shared Models." Artificial Intelligence and Statistics, 2024.

Markdown

[Cousins et al. "To Pool or Not to Pool: Analyzing the Regularizing Effects of Group-Fair Training on Shared Models." Artificial Intelligence and Statistics, 2024.](https://mlanthology.org/aistats/2024/cousins2024aistats-pool/)

BibTeX

@inproceedings{cousins2024aistats-pool,
  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 = {Artificial Intelligence and Statistics},
  year      = {2024},
  pages     = {4573-4581},
  volume    = {238},
  url       = {https://mlanthology.org/aistats/2024/cousins2024aistats-pool/}
}