{"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/path-sgd-path-normalized-optimization-in-deep","title":"Path-SGD: Path-Normalized Optimization in Deep Neural Networks","arxiv_id":"1506.02617","date":"2015-06-08","proceeding":"NeurIPS 2015 12","authors":["Behnam Neyshabur","Ruslan Salakhutdinov","Nathan Srebro"],"abstract":"We revisit the choice of SGD for training deep neural networks by\nreconsidering the appropriate geometry in which to optimize the weights. We\nargue for a geometry invariant to rescaling of weights that does not affect the\noutput of the network, and suggest Path-SGD, which is an approximate steepest\ndescent method with respect to a path-wise regularizer related to max-norm\nregularization. Path-SGD is easy and efficient to implement and leads to\nempirical gains over SGD and AdaGrad.","url_abs":"http://arxiv.org/abs/1506.02617v1","url_pdf":"http://arxiv.org/pdf/1506.02617v1.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":"path-sgd-path-normalized-optimization-in-deep","repo_url":"https://github.com/bneyshabur/path-sgd","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[],"methods":[{"method_slug":"adagrad","method_name":"AdaGrad"},{"method_slug":"sgd","method_name":"SGD"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1506.02617","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}