Learning Fair Representations for Kernel Models

Zilong Tan, Samuel Yeom, Matt Fredrikson, Ameet Talwalkar
Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, PMLR 108:155-166, 2020.

Abstract

Fair representations are a powerful tool for establishing criteria like statistical parity, proxy non-discrimination, and equality of opportunity in learned models. Existing techniques for learning these representations are typically model-agnostic, as they preprocess the original data such that the output satisfies some fairness criterion, and can be used with arbitrary learning methods. In contrast, we demonstrate the promise of learning a model-aware fair representation, focusing on kernel-based models. We leverage the classical Sufficient Dimension Reduction (SDR) framework to construct representations as subspaces of the reproducing kernel Hilbert space (RKHS), whose member functions are guaranteed to satisfy fairness. Our method supports several fairness criteria, continuous and discrete data, and multiple protected attributes. We further show how to calibrate the accuracy tradeoff by characterizing it in terms of the principal angles between subspaces of the RKHS. Finally, we apply our approach to obtain the first Fair Gaussian Process (FGP) prior for fair Bayesian learning, and show that it is competitive with, and in some cases outperforms, state-of-the-art methods on real data.

Cite this Paper


BibTeX
@InProceedings{pmlr-v108-tan20a, title = {Learning Fair Representations for Kernel Models}, author = {Tan, Zilong and Yeom, Samuel and Fredrikson, Matt and Talwalkar, Ameet}, booktitle = {Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics}, pages = {155--166}, year = {2020}, editor = {Chiappa, Silvia and Calandra, Roberto}, volume = {108}, series = {Proceedings of Machine Learning Research}, month = {26--28 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v108/tan20a/tan20a.pdf}, url = {http://proceedings.mlr.press/v108/tan20a.html}, abstract = {Fair representations are a powerful tool for establishing criteria like statistical parity, proxy non-discrimination, and equality of opportunity in learned models. Existing techniques for learning these representations are typically model-agnostic, as they preprocess the original data such that the output satisfies some fairness criterion, and can be used with arbitrary learning methods. In contrast, we demonstrate the promise of learning a model-aware fair representation, focusing on kernel-based models. We leverage the classical Sufficient Dimension Reduction (SDR) framework to construct representations as subspaces of the reproducing kernel Hilbert space (RKHS), whose member functions are guaranteed to satisfy fairness. Our method supports several fairness criteria, continuous and discrete data, and multiple protected attributes. We further show how to calibrate the accuracy tradeoff by characterizing it in terms of the principal angles between subspaces of the RKHS. Finally, we apply our approach to obtain the first Fair Gaussian Process (FGP) prior for fair Bayesian learning, and show that it is competitive with, and in some cases outperforms, state-of-the-art methods on real data.} }
Endnote
%0 Conference Paper %T Learning Fair Representations for Kernel Models %A Zilong Tan %A Samuel Yeom %A Matt Fredrikson %A Ameet Talwalkar %B Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2020 %E Silvia Chiappa %E Roberto Calandra %F pmlr-v108-tan20a %I PMLR %P 155--166 %U http://proceedings.mlr.press/v108/tan20a.html %V 108 %X Fair representations are a powerful tool for establishing criteria like statistical parity, proxy non-discrimination, and equality of opportunity in learned models. Existing techniques for learning these representations are typically model-agnostic, as they preprocess the original data such that the output satisfies some fairness criterion, and can be used with arbitrary learning methods. In contrast, we demonstrate the promise of learning a model-aware fair representation, focusing on kernel-based models. We leverage the classical Sufficient Dimension Reduction (SDR) framework to construct representations as subspaces of the reproducing kernel Hilbert space (RKHS), whose member functions are guaranteed to satisfy fairness. Our method supports several fairness criteria, continuous and discrete data, and multiple protected attributes. We further show how to calibrate the accuracy tradeoff by characterizing it in terms of the principal angles between subspaces of the RKHS. Finally, we apply our approach to obtain the first Fair Gaussian Process (FGP) prior for fair Bayesian learning, and show that it is competitive with, and in some cases outperforms, state-of-the-art methods on real data.
APA
Tan, Z., Yeom, S., Fredrikson, M. & Talwalkar, A.. (2020). Learning Fair Representations for Kernel Models. Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 108:155-166 Available from http://proceedings.mlr.press/v108/tan20a.html.

Related Material