{"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/deep-defense-training-dnns-with-improved","title":"Deep Defense: Training DNNs with Improved Adversarial Robustness","arxiv_id":"1803.00404","date":"2018-02-23","proceeding":"NeurIPS 2018 12","authors":["Ziang Yan","Yiwen Guo","Chang-Shui Zhang"],"abstract":"Despite the efficacy on a variety of computer vision tasks, deep neural\nnetworks (DNNs) are vulnerable to adversarial attacks, limiting their\napplications in security-critical systems. Recent works have shown the\npossibility of generating imperceptibly perturbed image inputs (a.k.a.,\nadversarial examples) to fool well-trained DNN classifiers into making\narbitrary predictions. To address this problem, we propose a training recipe\nnamed \"deep defense\". Our core idea is to integrate an adversarial\nperturbation-based regularizer into the classification objective, such that the\nobtained models learn to resist potential attacks, directly and precisely. The\nwhole optimization problem is solved just like training a recursive network.\nExperimental results demonstrate that our method outperforms training with\nadversarial/Parseval regularizations by large margins on various datasets\n(including MNIST, CIFAR-10 and ImageNet) and different DNN architectures. Code\nand models for reproducing our results are available at\nhttps://github.com/ZiangYan/deepdefense.pytorch","url_abs":"http://arxiv.org/abs/1803.00404v3","url_pdf":"http://arxiv.org/pdf/1803.00404v3.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":"deep-defense-training-dnns-with-improved","repo_url":"https://github.com/ZiangYan/deepdefense.pytorch","is_official":1,"mentioned_in_paper":1,"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":{"syntology_url":"https://syntology.ai/paper/1803.00404","atlas_url":"https://app.syntology.ai/?focus=1803.00404","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}