Adversarial examples are not bugs, they are features
© 2019 Neural information processing systems foundation. All rights reserved. Adversarial examples have attracted significant attention in machine learning, but the reasons for their existence and pervasiveness remain unclear. We demonstrate that adversarial examples can be directly attributed to th...
| Main Authors: | , , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
2021
|
| Online Access: | https://hdl.handle.net/1721.1/137500 |
| _version_ | 1811090467370041344 |
|---|---|
| author | Ilyas, A Santurkar, S Tsipras, D Engstrom, L Tran, B Madry, A |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Ilyas, A Santurkar, S Tsipras, D Engstrom, L Tran, B Madry, A |
| author_sort | Ilyas, A |
| collection | MIT |
| description | © 2019 Neural information processing systems foundation. All rights reserved. Adversarial examples have attracted significant attention in machine learning, but the reasons for their existence and pervasiveness remain unclear. We demonstrate that adversarial examples can be directly attributed to the presence of non-robust features: features (derived from patterns in the data distribution) that are highly predictive, yet brittle and (thus) incomprehensible to humans. After capturing these features within a theoretical framework, we establish their widespread existence in standard datasets. Finally, we present a simple setting where we can rigorously tie the phenomena we observe in practice to a misalignment between the (human-specified) notion of robustness and the inherent geometry of the data. |
| first_indexed | 2024-09-23T14:46:38Z |
| format | Article |
| id | mit-1721.1/137500 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T14:46:38Z |
| publishDate | 2021 |
| record_format | dspace |
| spelling | mit-1721.1/1375002023-02-06T20:59:21Z Adversarial examples are not bugs, they are features Ilyas, A Santurkar, S Tsipras, D Engstrom, L Tran, B Madry, A Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science © 2019 Neural information processing systems foundation. All rights reserved. Adversarial examples have attracted significant attention in machine learning, but the reasons for their existence and pervasiveness remain unclear. We demonstrate that adversarial examples can be directly attributed to the presence of non-robust features: features (derived from patterns in the data distribution) that are highly predictive, yet brittle and (thus) incomprehensible to humans. After capturing these features within a theoretical framework, we establish their widespread existence in standard datasets. Finally, we present a simple setting where we can rigorously tie the phenomena we observe in practice to a misalignment between the (human-specified) notion of robustness and the inherent geometry of the data. 2021-11-05T15:00:20Z 2021-11-05T15:00:20Z 2019 2021-02-02T14:05:36Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/137500 Ilyas, A, Santurkar, S, Tsipras, D, Engstrom, L, Tran, B et al. 2019. "Adversarial examples are not bugs, they are features." Advances in Neural Information Processing Systems, 32. en https://papers.nips.cc/paper/2019/hash/e2c420d928d4bf8ce0ff2ec19b371514-Abstract.html Advances in Neural Information Processing Systems Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. application/pdf Neural Information Processing Systems (NIPS) |
| spellingShingle | Ilyas, A Santurkar, S Tsipras, D Engstrom, L Tran, B Madry, A Adversarial examples are not bugs, they are features |
| title | Adversarial examples are not bugs, they are features |
| title_full | Adversarial examples are not bugs, they are features |
| title_fullStr | Adversarial examples are not bugs, they are features |
| title_full_unstemmed | Adversarial examples are not bugs, they are features |
| title_short | Adversarial examples are not bugs, they are features |
| title_sort | adversarial examples are not bugs they are features |
| url | https://hdl.handle.net/1721.1/137500 |
| work_keys_str_mv | AT ilyasa adversarialexamplesarenotbugstheyarefeatures AT santurkars adversarialexamplesarenotbugstheyarefeatures AT tsiprasd adversarialexamplesarenotbugstheyarefeatures AT engstroml adversarialexamplesarenotbugstheyarefeatures AT tranb adversarialexamplesarenotbugstheyarefeatures AT madrya adversarialexamplesarenotbugstheyarefeatures |