{"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/disentangling-adversarial-robustness-and","title":"Disentangling Adversarial Robustness and Generalization","arxiv_id":"1812.00740","date":"2018-12-03","proceeding":"CVPR 2019 6","authors":["David Stutz","Matthias Hein","Bernt Schiele"],"abstract":"Obtaining deep networks that are robust against adversarial examples and\ngeneralize well is an open problem. A recent hypothesis even states that both\nrobust and accurate models are impossible, i.e., adversarial robustness and\ngeneralization are conflicting goals. In an effort to clarify the relationship\nbetween robustness and generalization, we assume an underlying, low-dimensional\ndata manifold and show that: 1. regular adversarial examples leave the\nmanifold; 2. adversarial examples constrained to the manifold, i.e.,\non-manifold adversarial examples, exist; 3. on-manifold adversarial examples\nare generalization errors, and on-manifold adversarial training boosts\ngeneralization; 4. regular robustness and generalization are not necessarily\ncontradicting goals. These assumptions imply that both robust and accurate\nmodels are possible. However, different models (architectures, training\nstrategies etc.) can exhibit different robustness and generalization\ncharacteristics. To confirm our claims, we present extensive experiments on\nsynthetic data (with known manifold) as well as on EMNIST, Fashion-MNIST and\nCelebA.","url_abs":"http://arxiv.org/abs/1812.00740v2","url_pdf":"http://arxiv.org/pdf/1812.00740v2.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":"disentangling-adversarial-robustness-and","repo_url":"https://github.com/davidstutz/cvpr2019-adversarial-robustness","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"disentangling-adversarial-robustness-and","repo_url":"https://github.com/davidstutz/disentangling-robustness-generalization","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"adversarial-robustness","task_name":"Adversarial Robustness"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1812.00740","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}