{"id":696811,"date":"2020-10-07T17:11:17","date_gmt":"2020-10-08T00:11:17","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-research-item&#038;p=696811"},"modified":"2022-11-16T14:53:04","modified_gmt":"2022-11-16T22:53:04","slug":"mace-a-flexible-framework-for-membership-privacy-estimation-in-generative-models","status":"publish","type":"msr-research-item","link":"https:\/\/www.microsoft.com\/en-us\/research\/publication\/mace-a-flexible-framework-for-membership-privacy-estimation-in-generative-models\/","title":{"rendered":"MACE: A Flexible Framework for Membership Privacy Estimation in Generative Models"},"content":{"rendered":"<p><span dir=\"ltr\" role=\"presentation\">Generative machine learning models are being increasingly viewed as a way to share sensitive <\/span><span dir=\"ltr\" role=\"presentation\">data between institutions. While there has been work on developing differentially private <\/span><span dir=\"ltr\" role=\"presentation\">generative modeling approaches, these approaches generally lead to sub-par sample quality, <\/span><span dir=\"ltr\" role=\"presentation\">limiting their use in real world applications. Another line of work has focused on developing <\/span><span dir=\"ltr\" role=\"presentation\">generative models which lead to higher quality samples but currently lack any formal <\/span><span dir=\"ltr\" role=\"presentation\">privacy guarantees. In this work, we propose the first formal framework for membership <\/span><span dir=\"ltr\" role=\"presentation\">privacy estimation in generative models. We formulate the membership privacy risk as a <\/span><span dir=\"ltr\" role=\"presentation\">statistical divergence between training samples and hold-out samples, and propose sample-<\/span><span dir=\"ltr\" role=\"presentation\">based methods to estimate this divergence. Compared to previous works, our framework <\/span><span dir=\"ltr\" role=\"presentation\">makes more realistic and flexible assumptions. First, we offer a generalizable metric as an <\/span><span dir=\"ltr\" role=\"presentation\">alternative to the accuracy metric (Yeom et al., 2018; Hayes et al., 2019) especially for <\/span><span dir=\"ltr\" role=\"presentation\">imbalanced datasets. Second, we loosen the assumption of having full access to the underlying <\/span><span dir=\"ltr\" role=\"presentation\">distribution from previous studies (Yeom et al., 2018; Jayaraman et al., 2020), and propose <\/span><span dir=\"ltr\" role=\"presentation\">sample-based estimations with theoretical guarantees. Third, along with the population-level <\/span><span dir=\"ltr\" role=\"presentation\">membership privacy risk estimation via the optimal membership advantage, we offer the <\/span><span dir=\"ltr\" role=\"presentation\">individual-level estimation via the individual privacy risk. Fourth, our framework allows <\/span><span dir=\"ltr\" role=\"presentation\">adversaries to access the trained model via a customized query, while prior works require <\/span><span dir=\"ltr\" role=\"presentation\">specific attributes (Hayes et al., 2019; Chen et al., 2019; Hilprecht et al., 2019).<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Generative machine learning models are being increasingly viewed as a way to share sensitive data between institutions. While there has been work on developing differentially private generative modeling approaches, these approaches generally lead to sub-par sample quality, limiting their use in real world applications. Another line of work has focused on developing generative models which [&hellip;]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"footnotes":""},"msr-content-type":[3],"msr-research-highlight":[],"research-area":[13556,13558],"msr-publication-type":[193715],"msr-product-type":[],"msr-focus-area":[],"msr-platform":[],"msr-download-source":[],"msr-locale":[268875],"msr-field-of-study":[],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-696811","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":"2022-10-6","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"Transactions on Machine Learning Research (TMLR)","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\/pdf?id=Zxm0kNe3u7","label_id":"243109","label":0}],"msr_related_uploader":"","msr_attachments":[],"msr-author-ordering":[{"type":"user_nicename","value":"Yixi Xu","user_id":39775,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Yixi Xu"},{"type":"user_nicename","value":"Sumit Mukherjee","user_id":39778,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Sumit Mukherjee"},{"type":"text","value":"Xiyang Liu","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Shruti Tople","user_id":39003,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Shruti Tople"},{"type":"user_nicename","value":"Rahul Dodhia","user_id":41401,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Rahul Dodhia"},{"type":"edited_text","value":"Juan M. Lavista Ferres","user_id":39552,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Juan M. Lavista Ferres"}],"msr_impact_theme":[],"msr_research_lab":[199561],"msr_event":[],"msr_group":[559983,696544],"msr_project":[778522,648207],"publication":[],"video":[],"download":[],"msr_publication_type":"article","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/696811"}],"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":4,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/696811\/revisions"}],"predecessor-version":[{"id":898998,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/696811\/revisions\/898998"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=696811"}],"wp:term":[{"taxonomy":"msr-content-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-content-type?post=696811"},{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=696811"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=696811"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=696811"},{"taxonomy":"msr-product-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-product-type?post=696811"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=696811"},{"taxonomy":"msr-platform","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-platform?post=696811"},{"taxonomy":"msr-download-source","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-download-source?post=696811"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=696811"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=696811"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=696811"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=696811"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=696811"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=696811"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}