{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/the-adversarial-attack-and-detection-under","title":"The Adversarial Attack and Detection under the Fisher Information Metric","arxiv_id":"1810.03806","date":"2018-10-09","proceeding":null,"authors":["Chenxiao Zhao","P. Thomas Fletcher","Mixue Yu","Yaxin Peng","Guixu Zhang","Chaomin Shen"],"abstract":"Many deep learning models are vulnerable to the adversarial attack, i.e.,\nimperceptible but intentionally-designed perturbations to the input can cause\nincorrect output of the networks. In this paper, using information geometry, we\nprovide a reasonable explanation for the vulnerability of deep learning models.\nBy considering the data space as a non-linear space with the Fisher information\nmetric induced from a neural network, we first propose an adversarial attack\nalgorithm termed one-step spectral attack (OSSA). The method is described by a\nconstrained quadratic form of the Fisher information matrix, where the optimal\nadversarial perturbation is given by the first eigenvector, and the model\nvulnerability is reflected by the eigenvalues. The larger an eigenvalue is, the\nmore vulnerable the model is to be attacked by the corresponding eigenvector.\nTaking advantage of the property, we also propose an adversarial detection\nmethod with the eigenvalues serving as characteristics. Both our attack and\ndetection algorithms are numerically optimized to work efficiently on large\ndatasets. Our evaluations show superior performance compared with other\nmethods, implying that the Fisher information is a promising approach to\ninvestigate the adversarial attacks and defenses.","url_abs":"http://arxiv.org/abs/1810.03806v2","url_pdf":"http://arxiv.org/pdf/1810.03806v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"the-adversarial-attack-and-detection-under","repo_url":"https://github.com/eliot-tron/curvnetattack","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"adversarial-attack","task_name":"Adversarial Attack"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1810.03806","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}