{"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/cnn-cert-an-efficient-framework-for","title":"CNN-Cert: An Efficient Framework for Certifying Robustness of Convolutional Neural Networks","arxiv_id":"1811.12395","date":"2018-11-29","proceeding":null,"authors":["Akhilan Boopathy","Tsui-Wei Weng","Pin-Yu Chen","Sijia Liu","Luca Daniel"],"abstract":"Verifying robustness of neural network classifiers has attracted great\ninterests and attention due to the success of deep neural networks and their\nunexpected vulnerability to adversarial perturbations. Although finding minimum\nadversarial distortion of neural networks (with ReLU activations) has been\nshown to be an NP-complete problem, obtaining a non-trivial lower bound of\nminimum distortion as a provable robustness guarantee is possible. However,\nmost previous works only focused on simple fully-connected layers (multilayer\nperceptrons) and were limited to ReLU activations. This motivates us to propose\na general and efficient framework, CNN-Cert, that is capable of certifying\nrobustness on general convolutional neural networks. Our framework is general\n-- we can handle various architectures including convolutional layers,\nmax-pooling layers, batch normalization layer, residual blocks, as well as\ngeneral activation functions; our approach is efficient -- by exploiting the\nspecial structure of convolutional layers, we achieve up to 17 and 11 times of\nspeed-up compared to the state-of-the-art certification algorithms (e.g.\nFast-Lin, CROWN) and 366 times of speed-up compared to the dual-LP approach\nwhile our algorithm obtains similar or even better verification bounds. In\naddition, CNN-Cert generalizes state-of-the-art algorithms e.g. Fast-Lin and\nCROWN. We demonstrate by extensive experiments that our method outperforms\nstate-of-the-art lower-bound-based certification algorithms in terms of both\nbound quality and speed.","url_abs":"http://arxiv.org/abs/1811.12395v1","url_pdf":"http://arxiv.org/pdf/1811.12395v1.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":"cnn-cert-an-efficient-framework-for","repo_url":"https://github.com/AkhilanB/CNN-Cert","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"cnn-cert-an-efficient-framework-for","repo_url":"https://github.com/ZhaoyangLyu/FROWN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[],"methods":[{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"relu","method_name":"ReLU"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1811.12395","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}