{"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-differentially-private-recurrent","title":"Learning Differentially Private Recurrent Language Models","arxiv_id":"1710.06963","date":"2017-10-18","proceeding":"ICLR 2018 1","authors":["H. Brendan McMahan","Daniel Ramage","Kunal Talwar","Li Zhang"],"abstract":"We demonstrate that it is possible to train large recurrent language models\nwith user-level differential privacy guarantees with only a negligible cost in\npredictive accuracy. Our work builds on recent advances in the training of deep\nnetworks on user-partitioned data and privacy accounting for stochastic\ngradient descent. In particular, we add user-level privacy protection to the\nfederated averaging algorithm, which makes \"large step\" updates from user-level\ndata. Our work demonstrates that given a dataset with a sufficiently large\nnumber of users (a requirement easily met by even small internet-scale\ndatasets), achieving differential privacy comes at the cost of increased\ncomputation, rather than in decreased utility as in most prior work. We find\nthat our private LSTM language models are quantitatively and qualitatively\nsimilar to un-noised models when trained on a large dataset.","url_abs":"http://arxiv.org/abs/1710.06963v3","url_pdf":"http://arxiv.org/pdf/1710.06963v3.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-differentially-private-recurrent","repo_url":"https://github.com/apple/ml-flair","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"NOASSERTION"}}],"tasks":[],"methods":[{"method_slug":"lstm","method_name":"LSTM"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"tanh-activation","method_name":"Tanh Activation"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1710.06963","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}