{"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/lipschitz-margin-training-scalable","title":"Lipschitz-Margin Training: Scalable Certification of Perturbation Invariance for Deep Neural Networks","arxiv_id":"1802.04034","date":"2018-02-12","proceeding":"NeurIPS 2018 12","authors":["Yusuke Tsuzuku","Issei Sato","Masashi Sugiyama"],"abstract":"High sensitivity of neural networks against malicious perturbations on inputs\ncauses security concerns. To take a steady step towards robust classifiers, we\naim to create neural network models provably defended from perturbations. Prior\ncertification work requires strong assumptions on network structures and\nmassive computational costs, and thus the range of their applications was\nlimited. From the relationship between the Lipschitz constants and prediction\nmargins, we present a computationally efficient calculation technique to\nlower-bound the size of adversarial perturbations that can deceive networks,\nand that is widely applicable to various complicated networks. Moreover, we\npropose an efficient training procedure that robustifies networks and\nsignificantly improves the provably guarded areas around data points. In\nexperimental evaluations, our method showed its ability to provide a\nnon-trivial guarantee and enhance robustness for even large networks.","url_abs":"http://arxiv.org/abs/1802.04034v3","url_pdf":"http://arxiv.org/pdf/1802.04034v3.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":"lipschitz-margin-training-scalable","repo_url":"https://github.com/ytsmiling/lmt","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}},{"paper_slug":"lipschitz-margin-training-scalable","repo_url":"https://github.com/sungyoon-lee/bcp","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1802.04034","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}