Evaluating Model Bias Requires Characterizing its Mistakes

Isabela Albuquerque, Jessica Schrouff, David Warde-Farley, Ali Taylan Cemgil, Sven Gowal, Olivia Wiles
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:938-954, 2024.

Abstract

The ability to properly benchmark model performance in the face of spurious correlations is important to both build better predictors and increase confidence that models are operating as intended. We demonstrate that characterizing (as opposed to simply quantifying) model mistakes across subgroups is pivotal to properly reflect model biases, which are ignored by standard metrics such as worst-group accuracy or accuracy gap. Inspired by the hypothesis testing framework, we introduce SkewSize, a principled and flexible metric that captures bias from mistakes in a model’s predictions. It can be used in multi-class settings or generalised to the open vocabulary setting of generative models. SkewSize is an aggregation of the effect size of the interaction between two categorical variables: the spurious variable representing the bias attribute the model’s prediction. We demonstrate the utility of SkewSize in multiple settings including: standard vision models trained on synthetic data, vision models trained on ImageNet, and large scale vision-and-language models from the BLIP-2 family. In each case, the proposed SkewSize is able to highlight biases not captured by other metrics, while also providing insights on the impact of recently proposed techniques, such as instruction tuning.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-albuquerque24a, title = {Evaluating Model Bias Requires Characterizing its Mistakes}, author = {Albuquerque, Isabela and Schrouff, Jessica and Warde-Farley, David and Cemgil, Ali Taylan and Gowal, Sven and Wiles, Olivia}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {938--954}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/albuquerque24a/albuquerque24a.pdf}, url = {https://proceedings.mlr.press/v235/albuquerque24a.html}, abstract = {The ability to properly benchmark model performance in the face of spurious correlations is important to both build better predictors and increase confidence that models are operating as intended. We demonstrate that characterizing (as opposed to simply quantifying) model mistakes across subgroups is pivotal to properly reflect model biases, which are ignored by standard metrics such as worst-group accuracy or accuracy gap. Inspired by the hypothesis testing framework, we introduce SkewSize, a principled and flexible metric that captures bias from mistakes in a model’s predictions. It can be used in multi-class settings or generalised to the open vocabulary setting of generative models. SkewSize is an aggregation of the effect size of the interaction between two categorical variables: the spurious variable representing the bias attribute the model’s prediction. We demonstrate the utility of SkewSize in multiple settings including: standard vision models trained on synthetic data, vision models trained on ImageNet, and large scale vision-and-language models from the BLIP-2 family. In each case, the proposed SkewSize is able to highlight biases not captured by other metrics, while also providing insights on the impact of recently proposed techniques, such as instruction tuning.} }
Endnote
%0 Conference Paper %T Evaluating Model Bias Requires Characterizing its Mistakes %A Isabela Albuquerque %A Jessica Schrouff %A David Warde-Farley %A Ali Taylan Cemgil %A Sven Gowal %A Olivia Wiles %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-albuquerque24a %I PMLR %P 938--954 %U https://proceedings.mlr.press/v235/albuquerque24a.html %V 235 %X The ability to properly benchmark model performance in the face of spurious correlations is important to both build better predictors and increase confidence that models are operating as intended. We demonstrate that characterizing (as opposed to simply quantifying) model mistakes across subgroups is pivotal to properly reflect model biases, which are ignored by standard metrics such as worst-group accuracy or accuracy gap. Inspired by the hypothesis testing framework, we introduce SkewSize, a principled and flexible metric that captures bias from mistakes in a model’s predictions. It can be used in multi-class settings or generalised to the open vocabulary setting of generative models. SkewSize is an aggregation of the effect size of the interaction between two categorical variables: the spurious variable representing the bias attribute the model’s prediction. We demonstrate the utility of SkewSize in multiple settings including: standard vision models trained on synthetic data, vision models trained on ImageNet, and large scale vision-and-language models from the BLIP-2 family. In each case, the proposed SkewSize is able to highlight biases not captured by other metrics, while also providing insights on the impact of recently proposed techniques, such as instruction tuning.
APA
Albuquerque, I., Schrouff, J., Warde-Farley, D., Cemgil, A.T., Gowal, S. & Wiles, O.. (2024). Evaluating Model Bias Requires Characterizing its Mistakes. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:938-954 Available from https://proceedings.mlr.press/v235/albuquerque24a.html.

Related Material