{"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-sample-detection-for-deep-neural","title":"Adversarial Sample Detection for Deep Neural Network through Model Mutation Testing","arxiv_id":"1812.05793","date":"2018-12-14","proceeding":null,"authors":["Jingyi Wang","Guoliang Dong","Jun Sun","Xinyu Wang","Peixin Zhang"],"abstract":"Deep neural networks (DNN) have been shown to be useful in a wide range of\napplications. However, they are also known to be vulnerable to adversarial\nsamples. By transforming a normal sample with some carefully crafted human\nimperceptible perturbations, even highly accurate DNN make wrong decisions.\nMultiple defense mechanisms have been proposed which aim to hinder the\ngeneration of such adversarial samples. However, a recent work show that most\nof them are ineffective. In this work, we propose an alternative approach to\ndetect adversarial samples at runtime. Our main observation is that adversarial\nsamples are much more sensitive than normal samples if we impose random\nmutations on the DNN. We thus first propose a measure of `sensitivity' and show\nempirically that normal samples and adversarial samples have distinguishable\nsensitivity. We then integrate statistical hypothesis testing and model\nmutation testing to check whether an input sample is likely to be normal or\nadversarial at runtime by measuring its sensitivity. We evaluated our approach\non the MNIST and CIFAR10 datasets. The results show that our approach detects\nadversarial samples generated by state-of-the-art attacking methods efficiently\nand accurately.","url_abs":"http://arxiv.org/abs/1812.05793v2","url_pdf":"http://arxiv.org/pdf/1812.05793v2.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-sample-detection-for-deep-neural","repo_url":"https://github.com/dgl-prc/m_testing_adversatial_sample","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":null},{"paper_slug":"adversarial-sample-detection-for-deep-neural","repo_url":"https://github.com/KuoTzu-yang/DeepMutation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"adversarial-sample-detection-for-deep-neural","repo_url":"https://github.com/KuoTzu-yang/DeepMutationOperators","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"adversarial-sample-detection-for-deep-neural","repo_url":"https://github.com/feiwu77777/Leveraging-Model-Interpretability-and-Stability-to-increase-Model-Robustness","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"adversarial-sample-detection-for-deep-neural","repo_url":"https://github.com/tkuo-tkuo/DeepMutationOperators","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"sensitivity","task_name":"Sensitivity"},{"task_slug":"hypothesis-testing","task_name":"Two-sample testing"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1812.05793","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}