{"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/from-adversarial-training-to-generative","title":"Rob-GAN: Generator, Discriminator, and Adversarial Attacker","arxiv_id":"1807.10454","date":"2018-07-27","proceeding":"CVPR 2019 6","authors":["Xuanqing Liu","Cho-Jui Hsieh"],"abstract":"We study two important concepts in adversarial deep learning---adversarial\ntraining and generative adversarial network (GAN). Adversarial training is the\ntechnique used to improve the robustness of discriminator by combining\nadversarial attacker and discriminator in the training phase. GAN is commonly\nused for image generation by jointly optimizing discriminator and generator. We\nshow these two concepts are indeed closely related and can be used to\nstrengthen each other---adding a generator to the adversarial training\nprocedure can improve the robustness of discriminators, and adding an\nadversarial attack to GAN training can improve the convergence speed and lead\nto better generators. Combining these two insights, we develop a framework\ncalled Rob-GAN to jointly optimize generator and discriminator in the presence\nof adversarial attacks---the generator generates fake images to fool\ndiscriminator; the adversarial attacker perturbs real images to fool the\ndiscriminator, and the discriminator wants to minimize loss under fake and\nadversarial images. Through this end-to-end training procedure, we are able to\nsimultaneously improve the convergence speed of GAN training, the quality of\nsynthetic images, and the robustness of discriminator under strong adversarial\nattacks. Experimental results demonstrate that the obtained classifier is more\nrobust than the state-of-the-art adversarial training approach, and the\ngenerator outperforms SN-GAN on ImageNet-143.","url_abs":"http://arxiv.org/abs/1807.10454v3","url_pdf":"http://arxiv.org/pdf/1807.10454v3.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":"from-adversarial-training-to-generative","repo_url":"https://github.com/xuanqing94/AdvGAN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":null},{"paper_slug":"from-adversarial-training-to-generative","repo_url":"https://github.com/xuanqing94/RobGAN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"adversarial-attack","task_name":"Adversarial Attack"},{"task_slug":null,"task_name":"Generative Adversarial Network"},{"task_slug":"image-generation","task_name":"Image Generation"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"speed","method_name":"SPEED"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}