{"id":864054,"date":"2022-08-22T16:41:21","date_gmt":"2022-08-22T23:41:21","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/"},"modified":"2022-08-22T16:41:21","modified_gmt":"2022-08-22T23:41:21","slug":"dp-transformers-training-transformer-models-with-differential-privacy","status":"publish","type":"msr-research-item","link":"https:\/\/www.microsoft.com\/en-us\/research\/publication\/dp-transformers-training-transformer-models-with-differential-privacy\/","title":{"rendered":"dp-transformers: Training transformer models with differential privacy"},"content":{"rendered":"<p><span data-contrast=\"auto\">Transformer models have recently taken the field of Natural Language Processing (NLP) by storm as large language models based on the transformer architecture have shown impressive performance across a wide range of applications. However, when investigating these models in terms of Responsible AI, a valid concern remains that privacy-preserving techniques must be properly applied when these models are trained with private data.<\/span><span data-ccp-props=\"{}\">\u00a0<\/span><\/p>\n<p><span data-contrast=\"auto\">Differential Privacy (DP) has become a gold standard definition of privacy that offers rigorous privacy guarantees to individuals while enabling learning from a population. Among a vast set of applications, training machine learning models with DP in particular has the potential to extract great value from private data while protecting privacy of the participants.<\/span><span data-ccp-props=\"{}\">\u00a0<\/span><\/p>\n<p><span data-contrast=\"auto\">Motivated by our recent <\/span><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/openreview.net\/forum?id=Q42f0dfjECO\"><span data-contrast=\"none\">work<\/span><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><span data-contrast=\"auto\">, we are releasing a repository for training transformer models with differential privacy. Our repository is based on integrating<\/span><span data-contrast=\"auto\"> the<\/span> <span data-contrast=\"none\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/opacus.ai\/\">Opacus<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> library<\/span><span data-contrast=\"auto\"> to<\/span><span data-contrast=\"auto\"> the<\/span> <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/huggingface.co\/\"><span data-contrast=\"none\">Hugging Face<\/span><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><span data-contrast=\"auto\"> platform. We aim to serve the privacy-preserving ML community in utilizing the state-of-the-art models while respecting the privacy of the individuals constituting what these models learn from.<\/span><span data-ccp-props=\"{}\">\u00a0<\/span><\/p>\n<p>Authors have equally contributed to this work.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Transformer models have recently taken the field of Natural Language Processing (NLP) by storm as large language models based on the transformer architecture have shown impressive performance across a wide range of applications. However, when investigating these models in terms of Responsible AI, a valid concern remains that privacy-preserving techniques must be properly applied when [&hellip;]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","footnotes":""},"msr-content-type":[3],"msr-research-highlight":[],"research-area":[13556,13558],"msr-publication-type":[193726],"msr-product-type":[],"msr-focus-area":[],"msr-platform":[],"msr-download-source":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-864054","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-artificial-intelligence","msr-research-area-security-privacy-cryptography","msr-locale-en_us"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2022-7-28","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":1,"msr_main_download":"","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"url","viewUrl":"false","id":"false","title":"https:\/\/openreview.net\/forum?id=r5IzgVAoIe5","label_id":"243109","label":0}],"msr_related_uploader":"","msr_attachments":[],"msr-author-ordering":[{"type":"user_nicename","value":"Huseyin Atahan Inan","user_id":40426,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Huseyin Atahan Inan"},{"type":"user_nicename","value":"Andre Manoel","user_id":40504,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Andre Manoel"},{"type":"user_nicename","value":"Lukas Wutschitz","user_id":38775,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Lukas Wutschitz"}],"msr_impact_theme":[],"msr_research_lab":[],"msr_event":[],"msr_group":[756487,761911,793670,1054512],"msr_project":[866259],"publication":[],"video":[],"download":[],"msr_publication_type":"unpublished","related_content":{"projects":[{"ID":866259,"post_title":"dp-transformers","post_name":"dp-transformers","post_type":"msr-project","post_date":"2022-08-02 10:37:42","post_modified":"2022-08-04 09:16:44","post_status":"publish","permalink":"https:\/\/www.microsoft.com\/en-us\/research\/project\/dp-transformers\/","post_excerpt":"Training transformer models with differential privacy Transformer models have recently taken the field of Natural Language Processing (NLP) by storm as large language models based on the transformer architecture have shown impressive performance across a wide range of applications. However, when investigating these models in terms of Responsible AI, a valid concern remains that privacy-preserving techniques must be properly applied when these models are trained with private data.&nbsp; Differential Privacy (DP) has become a gold&hellip;","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/866259"}]}}]},"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/864054","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":7,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/864054\/revisions"}],"predecessor-version":[{"id":865914,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/864054\/revisions\/865914"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=864054"}],"wp:term":[{"taxonomy":"msr-content-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-content-type?post=864054"},{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=864054"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=864054"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=864054"},{"taxonomy":"msr-product-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-product-type?post=864054"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=864054"},{"taxonomy":"msr-platform","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-platform?post=864054"},{"taxonomy":"msr-download-source","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-download-source?post=864054"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=864054"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=864054"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=864054"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=864054"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=864054"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=864054"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=864054"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}