{"id":1152379,"date":"2025-10-17T08:35:56","date_gmt":"2025-10-17T15:35:56","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-video&#038;p=1152379"},"modified":"2025-10-17T08:35:57","modified_gmt":"2025-10-17T15:35:57","slug":"efficient-secure-aggregation-for-federated-learning","status":"publish","type":"msr-video","link":"https:\/\/www.microsoft.com\/en-us\/research\/video\/efficient-secure-aggregation-for-federated-learning\/","title":{"rendered":"Efficient Secure Aggregation for Federated Learning"},"content":{"rendered":"\n<p>Federated Learning\u202f(FL) trains a global model by having each selected device push only its model update to a central server, keeping raw data local. However, those updates can still leak sensitive information unless the server learns only their sum. A na\u00efve approach is to run a generic secure\u2011multiparty sum, but off\u2011the\u2011shelf protocols require several rounds of interaction and even direct client\u2011to\u2011client communication &#8211; often infeasible in FL, where mobile devices are intermittently online and can drop out at any moment, and cannot be expected to interact with each other.<\/p>\n\n\n\n<p>In this talk, I will review the secure\u2011aggregation problem in the context of FL and explain why na\u00efve solutions fail by focusing on constraints unique to the FL setting. I will then present Tacita, a single\u2011server protocol that satisfies these FL\u2011specific constraints while retaining provable security. Tacita uses an external committee (needed to prevent residual leakage) to aid in secure aggregation and offers:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>One\u2011shot execution: every client and every committee member sends exactly one message.<\/li>\n\n\n\n<li>Constant\u2011size communication per client, independent of the round\u2019s cohort of clients or committee size.<\/li>\n\n\n\n<li>Robust aggregation despite client or committee dropout.<\/li>\n<\/ul>\n\n\n\n<p>These properties are enabled by two new primitives: (i) succinct multi\u2011key linearly homomorphic threshold signatures\u202f(MKLHTS) for verifiable input soundness with a single aggregate signature, and (ii) a homomorphic variant of Silent Threshold Encryption\u202f(CRYPTO\u202f\u201924).<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Federated Learning\u202f(FL) trains a global model by having each selected device push only its model update to a central server, keeping raw data local. However, those updates can still leak sensitive information unless the server learns only their sum. A na\u00efve approach is to run a generic secure\u2011multiparty sum, but off\u2011the\u2011shelf protocols require several rounds [&hellip;]<\/p>\n","protected":false},"featured_media":1152380,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr_hide_image_in_river":null,"footnotes":""},"research-area":[13558],"msr-video-type":[269676],"msr-locale":[268875],"msr-post-option":[],"msr-session-type":[],"msr-impact-theme":[],"msr-pillar":[],"msr-episode":[],"msr-research-theme":[],"class_list":["post-1152379","msr-video","type-msr-video","status-publish","has-post-thumbnail","hentry","msr-research-area-security-privacy-cryptography","msr-video-type-cryptography-talk-series","msr-locale-en_us"],"msr_download_urls":"","msr_external_url":"https:\/\/youtu.be\/V1HH58z3dwQ","msr_secondary_video_url":"","msr_video_file":"http:\/\/0","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1152379","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-video"}],"version-history":[{"count":1,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1152379\/revisions"}],"predecessor-version":[{"id":1152382,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1152379\/revisions\/1152382"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/1152380"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1152379"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1152379"},{"taxonomy":"msr-video-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video-type?post=1152379"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1152379"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1152379"},{"taxonomy":"msr-session-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-session-type?post=1152379"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1152379"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1152379"},{"taxonomy":"msr-episode","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-episode?post=1152379"},{"taxonomy":"msr-research-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-theme?post=1152379"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}