{"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/learning-with-symmetric-label-noise-the","title":"Learning with Symmetric Label Noise: The Importance of Being Unhinged","arxiv_id":"1505.07634","date":"2015-05-28","proceeding":"NeurIPS 2015 12","authors":["Brendan van Rooyen","Aditya Krishna Menon","Robert C. Williamson"],"abstract":"Convex potential minimisation is the de facto approach to binary\nclassification. However, Long and Servedio [2010] proved that under symmetric\nlabel noise (SLN), minimisation of any convex potential over a linear function\nclass can result in classification performance equivalent to random guessing.\nThis ostensibly shows that convex losses are not SLN-robust. In this paper, we\npropose a convex, classification-calibrated loss and prove that it is\nSLN-robust. The loss avoids the Long and Servedio [2010] result by virtue of\nbeing negatively unbounded. The loss is a modification of the hinge loss, where\none does not clamp at zero; hence, we call it the unhinged loss. We show that\nthe optimal unhinged solution is equivalent to that of a strongly regularised\nSVM, and is the limiting solution for any convex potential; this implies that\nstrong l2 regularisation makes most standard learners SLN-robust. Experiments\nconfirm the SLN-robustness of the unhinged loss.","url_abs":"http://arxiv.org/abs/1505.07634v1","url_pdf":"http://arxiv.org/pdf/1505.07634v1.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":"learning-with-symmetric-label-noise-the","repo_url":"https://github.com/dmizr/phuber","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"binary-classification","task_name":"Binary Classification"},{"task_slug":"classification-1","task_name":"Classification"},{"task_slug":"classification","task_name":"General Classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1505.07634","atlas_url":"https://app.syntology.ai/?focus=1505.07634","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}