{"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/adversarial-playground-a-visualization-suite","title":"Adversarial-Playground: A Visualization Suite Showing How Adversarial Examples Fool Deep Learning","arxiv_id":"1708.00807","date":"2017-08-01","proceeding":null,"authors":["Andrew P. Norton","Yanjun Qi"],"abstract":"Recent studies have shown that attackers can force deep learning models to\nmisclassify so-called \"adversarial examples\": maliciously generated images\nformed by making imperceptible modifications to pixel values. With growing\ninterest in deep learning for security applications, it is important for\nsecurity experts and users of machine learning to recognize how learning\nsystems may be attacked. Due to the complex nature of deep learning, it is\nchallenging to understand how deep models can be fooled by adversarial\nexamples. Thus, we present a web-based visualization tool,\nAdversarial-Playground, to demonstrate the efficacy of common adversarial\nmethods against a convolutional neural network (CNN) system.\nAdversarial-Playground is educational, modular and interactive. (1) It enables\nnon-experts to compare examples visually and to understand why an adversarial\nexample can fool a CNN-based image classifier. (2) It can help security experts\nexplore more vulnerability of deep learning as a software module. (3) Building\nan interactive visualization is challenging in this domain due to the large\nfeature space of image classification (generating adversarial examples is slow\nin general and visualizing images are costly). Through multiple novel design\nchoices, our tool can provide fast and accurate responses to user requests.\nEmpirically, we find that our client-server division strategy reduced the\nresponse time by an average of 1.5 seconds per sample. Our other innovation, a\nfaster variant of JSMA evasion algorithm, empirically performed twice as fast\nas JSMA and yet maintains a comparable evasion rate.\n  Project source code and data from our experiments available at:\nhttps://github.com/QData/AdversarialDNN-Playground","url_abs":"http://arxiv.org/abs/1708.00807v1","url_pdf":"http://arxiv.org/pdf/1708.00807v1.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":"adversarial-playground-a-visualization-suite","repo_url":"https://github.com/QData/AdversarialDNN-Playground","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"adversarial-attack","task_name":"Adversarial Attack"},{"task_slug":"adversarial-defense","task_name":"Adversarial Defense"},{"task_slug":"deep-learning","task_name":"Deep Learning"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}