{"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/generalizable-adversarial-training-via","title":"Generalizable Adversarial Training via Spectral Normalization","arxiv_id":"1811.07457","date":"2018-11-19","proceeding":"ICLR 2019 5","authors":["Farzan Farnia","Jesse M. Zhang","David Tse"],"abstract":"Deep neural networks (DNNs) have set benchmarks on a wide array of supervised\nlearning tasks. Trained DNNs, however, often lack robustness to minor\nadversarial perturbations to the input, which undermines their true\npracticality. Recent works have increased the robustness of DNNs by fitting\nnetworks using adversarially-perturbed training samples, but the improved\nperformance can still be far below the performance seen in non-adversarial\nsettings. A significant portion of this gap can be attributed to the decrease\nin generalization performance due to adversarial training. In this work, we\nextend the notion of margin loss to adversarial settings and bound the\ngeneralization error for DNNs trained under several well-known gradient-based\nattack schemes, motivating an effective regularization scheme based on spectral\nnormalization of the DNN's weight matrices. We also provide a\ncomputationally-efficient method for normalizing the spectral norm of\nconvolutional layers with arbitrary stride and padding schemes in deep\nconvolutional networks. We evaluate the power of spectral normalization\nextensively on combinations of datasets, network architectures, and adversarial\ntraining schemes. The code is available at\nhttps://github.com/jessemzhang/dl_spectral_normalization.","url_abs":"http://arxiv.org/abs/1811.07457v1","url_pdf":"http://arxiv.org/pdf/1811.07457v1.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":"generalizable-adversarial-training-via","repo_url":"https://github.com/jessemzhang/dl_spectral_normalization","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1811.07457","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}