Feed Two Birds with One Scone: Exploiting Wild Data for Both Out-of-Distribution Generalization and Detection

Haoyue Bai, Gregory Canal, Xuefeng Du, Jeongyeol Kwon, Robert D Nowak, Yixuan Li
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:1454-1471, 2023.

Abstract

Modern machine learning models deployed in the wild can encounter both covariate and semantic shifts, giving rise to the problems of out-of-distribution (OOD) generalization and OOD detection respectively. While both problems have received significant research attention lately, they have been pursued independently. This may not be surprising, since the two tasks have seemingly conflicting goals. This paper provides a new unified approach that is capable of simultaneously generalizing to covariate shifts while robustly detecting semantic shifts. We propose a margin-based learning framework that exploits freely available unlabeled data in the wild that captures the environmental test-time OOD distributions under both covariate and semantic shifts. We show both empirically and theoretically that the proposed margin constraint is the key to achieving both OOD generalization and detection. Extensive experiments show the superiority of our framework, outperforming competitive baselines that specialize in either OOD generalization or OOD detection. Code is publicly available at https://github.com/deeplearning-wisc/scone.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-bai23a, title = {Feed Two Birds with One Scone: Exploiting Wild Data for Both Out-of-Distribution Generalization and Detection}, author = {Bai, Haoyue and Canal, Gregory and Du, Xuefeng and Kwon, Jeongyeol and Nowak, Robert D and Li, Yixuan}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {1454--1471}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/bai23a/bai23a.pdf}, url = {https://proceedings.mlr.press/v202/bai23a.html}, abstract = {Modern machine learning models deployed in the wild can encounter both covariate and semantic shifts, giving rise to the problems of out-of-distribution (OOD) generalization and OOD detection respectively. While both problems have received significant research attention lately, they have been pursued independently. This may not be surprising, since the two tasks have seemingly conflicting goals. This paper provides a new unified approach that is capable of simultaneously generalizing to covariate shifts while robustly detecting semantic shifts. We propose a margin-based learning framework that exploits freely available unlabeled data in the wild that captures the environmental test-time OOD distributions under both covariate and semantic shifts. We show both empirically and theoretically that the proposed margin constraint is the key to achieving both OOD generalization and detection. Extensive experiments show the superiority of our framework, outperforming competitive baselines that specialize in either OOD generalization or OOD detection. Code is publicly available at https://github.com/deeplearning-wisc/scone.} }
Endnote
%0 Conference Paper %T Feed Two Birds with One Scone: Exploiting Wild Data for Both Out-of-Distribution Generalization and Detection %A Haoyue Bai %A Gregory Canal %A Xuefeng Du %A Jeongyeol Kwon %A Robert D Nowak %A Yixuan Li %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-bai23a %I PMLR %P 1454--1471 %U https://proceedings.mlr.press/v202/bai23a.html %V 202 %X Modern machine learning models deployed in the wild can encounter both covariate and semantic shifts, giving rise to the problems of out-of-distribution (OOD) generalization and OOD detection respectively. While both problems have received significant research attention lately, they have been pursued independently. This may not be surprising, since the two tasks have seemingly conflicting goals. This paper provides a new unified approach that is capable of simultaneously generalizing to covariate shifts while robustly detecting semantic shifts. We propose a margin-based learning framework that exploits freely available unlabeled data in the wild that captures the environmental test-time OOD distributions under both covariate and semantic shifts. We show both empirically and theoretically that the proposed margin constraint is the key to achieving both OOD generalization and detection. Extensive experiments show the superiority of our framework, outperforming competitive baselines that specialize in either OOD generalization or OOD detection. Code is publicly available at https://github.com/deeplearning-wisc/scone.
APA
Bai, H., Canal, G., Du, X., Kwon, J., Nowak, R.D. & Li, Y.. (2023). Feed Two Birds with One Scone: Exploiting Wild Data for Both Out-of-Distribution Generalization and Detection. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:1454-1471 Available from https://proceedings.mlr.press/v202/bai23a.html.

Related Material