Exploring the landscape of spatial robustness
Copyright 2019 by the author(s). The study of adversarial robustness has so far largely focused on perturbations bound in lvnorms. However, state-of-the-art models turn out to be also vulnerable to other, more natural classes of perturbations such as translations and rotations. In this work, we thor...
| Main Authors: | , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
MLResearch Press
2021
|
| Online Access: | https://hdl.handle.net/1721.1/130391 |
| _version_ | 1826212230858801152 |
|---|---|
| author | Engstrom, Logan G. Tran, Brandon Tsipras, Dimitris Schmidt, Ludwig Madry, Aleksander |
| author2 | Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science |
| author_facet | Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Engstrom, Logan G. Tran, Brandon Tsipras, Dimitris Schmidt, Ludwig Madry, Aleksander |
| author_sort | Engstrom, Logan G. |
| collection | MIT |
| description | Copyright 2019 by the author(s). The study of adversarial robustness has so far largely focused on perturbations bound in lvnorms. However, state-of-the-art models turn out to be also vulnerable to other, more natural classes of perturbations such as translations and rotations. In this work, we thoroughly investigate the vulnerability of neural network-based classifiers to rotations and translations. While data augmentation offers relatively small robustness, we use ideas from robust optimization and test-time input aggregation to significantly improve robustness. Finally we find that, in contrast to the ip-norm case, first-order methods cannot reliably find worst-case perturbations. This highlights spatial robustness as a fundamentally different setting requiring additional study. |
| first_indexed | 2024-09-23T15:18:25Z |
| format | Article |
| id | mit-1721.1/130391 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T15:18:25Z |
| publishDate | 2021 |
| publisher | MLResearch Press |
| record_format | dspace |
| spelling | mit-1721.1/1303912022-09-29T14:05:04Z Exploring the landscape of spatial robustness Engstrom, Logan G. Tran, Brandon Tsipras, Dimitris Schmidt, Ludwig Madry, Aleksander Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Copyright 2019 by the author(s). The study of adversarial robustness has so far largely focused on perturbations bound in lvnorms. However, state-of-the-art models turn out to be also vulnerable to other, more natural classes of perturbations such as translations and rotations. In this work, we thoroughly investigate the vulnerability of neural network-based classifiers to rotations and translations. While data augmentation offers relatively small robustness, we use ideas from robust optimization and test-time input aggregation to significantly improve robustness. Finally we find that, in contrast to the ip-norm case, first-order methods cannot reliably find worst-case perturbations. This highlights spatial robustness as a fundamentally different setting requiring additional study. NSF (Grants CCF-1553428, CNS-1413920, CCF-1553428 and CNS-1815221) 2021-04-06T15:52:40Z 2021-04-06T15:52:40Z 2019-06 2021-02-05T18:20:48Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/130391 Engstrom, Logan et al. "Exploring the landscape of spatial robustness." Proceedings of the 36th International Conference on Machine Learning, June 2019, Long Beach, California, MLResearch Press, June 2019. © 2019 The Authors en http://proceedings.mlr.press/v97/engstrom19a.html Proceedings of the 36th International Conference on Machine Learning Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf MLResearch Press Proceedings of Machine Learning Research |
| spellingShingle | Engstrom, Logan G. Tran, Brandon Tsipras, Dimitris Schmidt, Ludwig Madry, Aleksander Exploring the landscape of spatial robustness |
| title | Exploring the landscape of spatial robustness |
| title_full | Exploring the landscape of spatial robustness |
| title_fullStr | Exploring the landscape of spatial robustness |
| title_full_unstemmed | Exploring the landscape of spatial robustness |
| title_short | Exploring the landscape of spatial robustness |
| title_sort | exploring the landscape of spatial robustness |
| url | https://hdl.handle.net/1721.1/130391 |
| work_keys_str_mv | AT engstromlogang exploringthelandscapeofspatialrobustness AT tranbrandon exploringthelandscapeofspatialrobustness AT tsiprasdimitris exploringthelandscapeofspatialrobustness AT schmidtludwig exploringthelandscapeofspatialrobustness AT madryaleksander exploringthelandscapeofspatialrobustness |