{"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/constructing-unrestricted-adversarial","title":"Constructing Unrestricted Adversarial Examples with Generative Models","arxiv_id":"1805.07894","date":"2018-05-21","proceeding":"NeurIPS 2018 12","authors":["Yang Song","Rui Shu","Nate Kushman","Stefano Ermon"],"abstract":"Adversarial examples are typically constructed by perturbing an existing data\npoint within a small matrix norm, and current defense methods are focused on\nguarding against this type of attack. In this paper, we propose unrestricted\nadversarial examples, a new threat model where the attackers are not restricted\nto small norm-bounded perturbations. Different from perturbation-based attacks,\nwe propose to synthesize unrestricted adversarial examples entirely from\nscratch using conditional generative models. Specifically, we first train an\nAuxiliary Classifier Generative Adversarial Network (AC-GAN) to model the\nclass-conditional distribution over data samples. Then, conditioned on a\ndesired class, we search over the AC-GAN latent space to find images that are\nlikely under the generative model and are misclassified by a target classifier.\nWe demonstrate through human evaluation that unrestricted adversarial examples\ngenerated this way are legitimate and belong to the desired class. Our\nempirical results on the MNIST, SVHN, and CelebA datasets show that\nunrestricted adversarial examples can bypass strong adversarial training and\ncertified defense methods designed for traditional adversarial attacks.","url_abs":"http://arxiv.org/abs/1805.07894v4","url_pdf":"http://arxiv.org/pdf/1805.07894v4.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":"constructing-unrestricted-adversarial","repo_url":"https://github.com/ermongroup/generative_adversary","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":null,"task_name":"Generative Adversarial Network"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1805.07894","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}