Improving Robustness against Real-World and Worst-Case Distribution Shifts through Decision Region Quantification

Leo Schwinn, Leon Bungert, An Nguyen, René Raab, Falk Pulsmeyer, Doina Precup, Bjoern Eskofier, Dario Zanca
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:19434-19449, 2022.

Abstract

The reliability of neural networks is essential for their use in safety-critical applications. Existing approaches generally aim at improving the robustness of neural networks to either real-world distribution shifts (e.g., common corruptions and perturbations, spatial transformations, and natural adversarial examples) or worst-case distribution shifts (e.g., optimized adversarial examples). In this work, we propose the Decision Region Quantification (DRQ) algorithm to improve the robustness of any differentiable pre-trained model against both real-world and worst-case distribution shifts in the data. DRQ analyzes the robustness of local decision regions in the vicinity of a given data point to make more reliable predictions. We theoretically motivate the DRQ algorithm by showing that it effectively smooths spurious local extrema in the decision surface. Furthermore, we propose an implementation using targeted and untargeted adversarial attacks. An extensive empirical evaluation shows that DRQ increases the robustness of adversarially and non-adversarially trained models against real-world and worst-case distribution shifts on several computer vision benchmark datasets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-schwinn22a, title = {Improving Robustness against Real-World and Worst-Case Distribution Shifts through Decision Region Quantification}, author = {Schwinn, Leo and Bungert, Leon and Nguyen, An and Raab, Ren{\'e} and Pulsmeyer, Falk and Precup, Doina and Eskofier, Bjoern and Zanca, Dario}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {19434--19449}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/schwinn22a/schwinn22a.pdf}, url = {https://proceedings.mlr.press/v162/schwinn22a.html}, abstract = {The reliability of neural networks is essential for their use in safety-critical applications. Existing approaches generally aim at improving the robustness of neural networks to either real-world distribution shifts (e.g., common corruptions and perturbations, spatial transformations, and natural adversarial examples) or worst-case distribution shifts (e.g., optimized adversarial examples). In this work, we propose the Decision Region Quantification (DRQ) algorithm to improve the robustness of any differentiable pre-trained model against both real-world and worst-case distribution shifts in the data. DRQ analyzes the robustness of local decision regions in the vicinity of a given data point to make more reliable predictions. We theoretically motivate the DRQ algorithm by showing that it effectively smooths spurious local extrema in the decision surface. Furthermore, we propose an implementation using targeted and untargeted adversarial attacks. An extensive empirical evaluation shows that DRQ increases the robustness of adversarially and non-adversarially trained models against real-world and worst-case distribution shifts on several computer vision benchmark datasets.} }
Endnote
%0 Conference Paper %T Improving Robustness against Real-World and Worst-Case Distribution Shifts through Decision Region Quantification %A Leo Schwinn %A Leon Bungert %A An Nguyen %A René Raab %A Falk Pulsmeyer %A Doina Precup %A Bjoern Eskofier %A Dario Zanca %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-schwinn22a %I PMLR %P 19434--19449 %U https://proceedings.mlr.press/v162/schwinn22a.html %V 162 %X The reliability of neural networks is essential for their use in safety-critical applications. Existing approaches generally aim at improving the robustness of neural networks to either real-world distribution shifts (e.g., common corruptions and perturbations, spatial transformations, and natural adversarial examples) or worst-case distribution shifts (e.g., optimized adversarial examples). In this work, we propose the Decision Region Quantification (DRQ) algorithm to improve the robustness of any differentiable pre-trained model against both real-world and worst-case distribution shifts in the data. DRQ analyzes the robustness of local decision regions in the vicinity of a given data point to make more reliable predictions. We theoretically motivate the DRQ algorithm by showing that it effectively smooths spurious local extrema in the decision surface. Furthermore, we propose an implementation using targeted and untargeted adversarial attacks. An extensive empirical evaluation shows that DRQ increases the robustness of adversarially and non-adversarially trained models against real-world and worst-case distribution shifts on several computer vision benchmark datasets.
APA
Schwinn, L., Bungert, L., Nguyen, A., Raab, R., Pulsmeyer, F., Precup, D., Eskofier, B. & Zanca, D.. (2022). Improving Robustness against Real-World and Worst-Case Distribution Shifts through Decision Region Quantification. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:19434-19449 Available from https://proceedings.mlr.press/v162/schwinn22a.html.

Related Material