{"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/efficient-formal-safety-analysis-of-neural","title":"Efficient Formal Safety Analysis of Neural Networks","arxiv_id":"1809.08098","date":"2018-09-19","proceeding":"NeurIPS 2018 12","authors":["Shiqi Wang","Kexin Pei","Justin Whitehouse","Junfeng Yang","Suman Jana"],"abstract":"Neural networks are increasingly deployed in real-world safety-critical\ndomains such as autonomous driving, aircraft collision avoidance, and malware\ndetection. However, these networks have been shown to often mispredict on\ninputs with minor adversarial or even accidental perturbations. Consequences of\nsuch errors can be disastrous and even potentially fatal as shown by the recent\nTesla autopilot crash. Thus, there is an urgent need for formal analysis\nsystems that can rigorously check neural networks for violations of different\nsafety properties such as robustness against adversarial perturbations within a\ncertain $L$-norm of a given image. An effective safety analysis system for a\nneural network must be able to either ensure that a safety property is\nsatisfied by the network or find a counterexample, i.e., an input for which the\nnetwork will violate the property. Unfortunately, most existing techniques for\nperforming such analysis struggle to scale beyond very small networks and the\nones that can scale to larger networks suffer from high false positives and\ncannot produce concrete counterexamples in case of a property violation. In\nthis paper, we present a new efficient approach for rigorously checking\ndifferent safety properties of neural networks that significantly outperforms\nexisting approaches by multiple orders of magnitude. Our approach can check\ndifferent safety properties and find concrete counterexamples for networks that\nare 10$\\times$ larger than the ones supported by existing analysis techniques.\nWe believe that our approach to estimating tight output bounds of a network for\na given input range can also help improve the explainability of neural networks\nand guide the training process of more robust neural networks.","url_abs":"http://arxiv.org/abs/1809.08098v3","url_pdf":"http://arxiv.org/pdf/1809.08098v3.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":"efficient-formal-safety-analysis-of-neural","repo_url":"https://github.com/tcwangshiqi-columbia/Neurify","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}},{"paper_slug":"efficient-formal-safety-analysis-of-neural","repo_url":"https://github.com/tcwangshiqi-columbia/ReluVal","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"NOASSERTION"}}],"tasks":[{"task_slug":"adversarial-attack","task_name":"Adversarial Attack"},{"task_slug":"adversarial-defense","task_name":"Adversarial Defense"},{"task_slug":"autonomous-driving","task_name":"Autonomous Driving"},{"task_slug":"collision-avoidance","task_name":"Collision Avoidance"},{"task_slug":"malware-detection","task_name":"Malware Detection"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1809.08098","atlas_url":"https://app.syntology.ai/?focus=1809.08098","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}