{"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/robust-loss-functions-under-label-noise-for","title":"Robust Loss Functions under Label Noise for Deep Neural Networks","arxiv_id":"1712.09482","date":"2017-12-27","proceeding":null,"authors":["Aritra Ghosh","Himanshu Kumar","P. S. Sastry"],"abstract":"In many applications of classifier learning, training data suffers from label\nnoise. Deep networks are learned using huge training data where the problem of\nnoisy labels is particularly relevant. The current techniques proposed for\nlearning deep networks under label noise focus on modifying the network\narchitecture and on algorithms for estimating true labels from noisy labels. An\nalternate approach would be to look for loss functions that are inherently\nnoise-tolerant. For binary classification there exist theoretical results on\nloss functions that are robust to label noise. In this paper, we provide some\nsufficient conditions on a loss function so that risk minimization under that\nloss function would be inherently tolerant to label noise for multiclass\nclassification problems. These results generalize the existing results on\nnoise-tolerant loss functions for binary classification. We study some of the\nwidely used loss functions in deep networks and show that the loss function\nbased on mean absolute value of error is inherently robust to label noise. Thus\nstandard back propagation is enough to learn the true classifier even under\nlabel noise. Through experiments, we illustrate the robustness of risk\nminimization with such loss functions for learning neural networks.","url_abs":"http://arxiv.org/abs/1712.09482v1","url_pdf":"http://arxiv.org/pdf/1712.09482v1.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":"robust-loss-functions-under-label-noise-for","repo_url":"https://github.com/smilelab-fl/fednoisy","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":{"atlas_url":"https://app.syntology.ai/?focus=1712.09482","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}