{"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/evading-defenses-to-transferable-adversarial","title":"Evading Defenses to Transferable Adversarial Examples by Translation-Invariant Attacks","arxiv_id":"1904.02884","date":"2019-04-05","proceeding":"CVPR 2019 6","authors":["Yinpeng Dong","Tianyu Pang","Hang Su","Jun Zhu"],"abstract":"Deep neural networks are vulnerable to adversarial examples, which can\nmislead classifiers by adding imperceptible perturbations. An intriguing\nproperty of adversarial examples is their good transferability, making\nblack-box attacks feasible in real-world applications. Due to the threat of\nadversarial attacks, many methods have been proposed to improve the robustness.\nSeveral state-of-the-art defenses are shown to be robust against transferable\nadversarial examples. In this paper, we propose a translation-invariant attack\nmethod to generate more transferable adversarial examples against the defense\nmodels. By optimizing a perturbation over an ensemble of translated images, the\ngenerated adversarial example is less sensitive to the white-box model being\nattacked and has better transferability. To improve the efficiency of attacks,\nwe further show that our method can be implemented by convolving the gradient\nat the untranslated image with a pre-defined kernel. Our method is generally\napplicable to any gradient-based attack method. Extensive experiments on the\nImageNet dataset validate the effectiveness of the proposed method. Our best\nattack fools eight state-of-the-art defenses at an 82% success rate on average\nbased only on the transferability, demonstrating the insecurity of the current\ndefense techniques.","url_abs":"http://arxiv.org/abs/1904.02884v1","url_pdf":"http://arxiv.org/pdf/1904.02884v1.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":"evading-defenses-to-transferable-adversarial","repo_url":"https://github.com/dongyp13/Translation-Invariant-Attacks","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"evading-defenses-to-transferable-adversarial","repo_url":"https://github.com/Trustworthy-AI-Group/TransferAttack","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"evading-defenses-to-transferable-adversarial","repo_url":"https://github.com/MindSpore-scientific-2/code-1/tree/main/Translation-Invariant/model","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null}],"tasks":[{"task_slug":"translation","task_name":"Translation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1904.02884","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}