{"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/poison-frogs-targeted-clean-label-poisoning","title":"Poison Frogs! Targeted Clean-Label Poisoning Attacks on Neural Networks","arxiv_id":"1804.00792","date":"2018-04-03","proceeding":"NeurIPS 2018 12","authors":["Ali Shafahi","W. Ronny Huang","Mahyar Najibi","Octavian Suciu","Christoph Studer","Tudor Dumitras","Tom Goldstein"],"abstract":"Data poisoning is an attack on machine learning models wherein the attacker\nadds examples to the training set to manipulate the behavior of the model at\ntest time. This paper explores poisoning attacks on neural nets. The proposed\nattacks use \"clean-labels\"; they don't require the attacker to have any control\nover the labeling of training data. They are also targeted; they control the\nbehavior of the classifier on a $\\textit{specific}$ test instance without\ndegrading overall classifier performance. For example, an attacker could add a\nseemingly innocuous image (that is properly labeled) to a training set for a\nface recognition engine, and control the identity of a chosen person at test\ntime. Because the attacker does not need to control the labeling function,\npoisons could be entered into the training set simply by leaving them on the\nweb and waiting for them to be scraped by a data collection bot.\n  We present an optimization-based method for crafting poisons, and show that\njust one single poison image can control classifier behavior when transfer\nlearning is used. For full end-to-end training, we present a \"watermarking\"\nstrategy that makes poisoning reliable using multiple ($\\approx$50) poisoned\ntraining instances. We demonstrate our method by generating poisoned frog\nimages from the CIFAR dataset and using them to manipulate image classifiers.","url_abs":"http://arxiv.org/abs/1804.00792v2","url_pdf":"http://arxiv.org/pdf/1804.00792v2.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":"poison-frogs-targeted-clean-label-poisoning","repo_url":"https://github.com/ashafahi/inceptionv3-transferLearn-poison","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}},{"paper_slug":"poison-frogs-targeted-clean-label-poisoning","repo_url":"https://github.com/JonasGeiping/poisoning-gradient-matching","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"poison-frogs-targeted-clean-label-poisoning","repo_url":"https://github.com/LostOxygen/poison_froggo","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"poison-frogs-targeted-clean-label-poisoning","repo_url":"https://github.com/zero-or-one/URP","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"poison-frogs-targeted-clean-label-poisoning","repo_url":"https://github.com/Recognito-Vision/Android-FaceRecognition-FaceLivenessDetection","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok"}}],"tasks":[{"task_slug":"data-poisoning","task_name":"Data Poisoning"},{"task_slug":"face-recognition","task_name":"Face Recognition"},{"task_slug":"transfer-learning","task_name":"Transfer Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1804.00792","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}