{"id":812314,"date":"2022-01-14T11:13:52","date_gmt":"2022-01-14T19:13:52","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-research-item&p=812314"},"modified":"2023-03-03T16:38:49","modified_gmt":"2023-03-04T00:38:49","slug":"matchmaker-data-drift-mitigation-in-machine-learning-for-large-scale-systemsmatchmaker-data-drift-mitigation-in-machine-learning-for-large-scale-systems","status":"publish","type":"msr-research-item","link":"https:\/\/www.microsoft.com\/en-us\/research\/publication\/matchmaker-data-drift-mitigation-in-machine-learning-for-large-scale-systemsmatchmaker-data-drift-mitigation-in-machine-learning-for-large-scale-systems\/","title":{"rendered":"Matchmaker: Data Drift Mitigation in Machine Learning for Large-Scale Systems"},"content":{"rendered":"

Today\u2019s data centers rely more heavily on machine learning (ML) in their deployed systems. However, these <\/span>systems are vulnerable to the<\/span> data drift<\/span> problem, that is, a mismatch between training data in the past and test data <\/span>in the future, which can lead to significant performance degradation and system inefficiencies. In this paper, we <\/span>demonstrate the impact of data drift in production by studying two real-world deployments in a leading cloud <\/span>provider. Our study shows that, despite frequent model retraining, these deployed models experience major <\/span>accuracy drops (up to 40%) and high accuracy variation, which lead to significant increase in operational costs. <\/span>Existing solutions to the data drift problem are not designed for large-scale deployments, which need to address <\/span>real-world issues such as scalability, ground truth latency, and mixed types of data drift. We propose<\/span> Matchmaker<\/span>, <\/span>the first scalable, adaptive, and flexible solution to the data drift problem in large-scale production systems. <\/span>Matchmaker<\/span> finds the most<\/span> similar<\/span> training data batch and uses the corresponding ML model for inference on <\/span>each<\/span> test point. As part of<\/span> Matchmaker<\/span>, we introduce a novel similarity metric to address multiple types of <\/span>data drifts while only incurring limited overhead. Experiments on our two real-world ML deployments show <\/span>that<\/span> Matchmaker<\/span> significantly improves model accuracy (up to 14% and 2%), which saves 18% and 1% in the <\/span>operational costs. At the same time,<\/span> Matchmaker<\/span> provides 8<\/span>\u00d7<\/span> and 4<\/span>\u00d7<\/span> faster predictions than a state-of-the-art <\/span>ML data drift solution, AUE.<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

Today\u2019s data centers rely more heavily on machine learning (ML) in their deployed systems. However, these systems are vulnerable to the data drift problem, that is, a mismatch between training data in the past and test data in the future, which can lead to significant performance degradation and system inefficiencies. In this paper, we demonstrate […]<\/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":"","msr-author-ordering":null,"msr_publishername":"","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"","msr_editors":"","msr_how_published":"","msr_isbn":"","msr_issue":"","msr_journal":"","msr_number":"","msr_organization":"","msr_pages_string":"","msr_page_range_start":"","msr_page_range_end":"","msr_series":"","msr_volume":"","msr_copyright":"","msr_conference_name":"MLSys","msr_doi":"","msr_arxiv_id":"","msr_s2_paper_id":"","msr_mag_id":"","msr_pubmed_id":"","msr_other_authors":"","msr_other_contributors":"","msr_speaker":"","msr_award":"","msr_affiliation":"","msr_institution":"","msr_host":"","msr_version":"","msr_duration":"","msr_original_fields_of_study":"","msr_release_tracker_id":"","msr_s2_match_type":"","msr_citation_count_updated":"","msr_published_date":"2022-1-1","msr_highlight_text":"","msr_notes":"","msr_longbiography":"","msr_publicationurl":"","msr_external_url":"","msr_secondary_video_url":"","msr_conference_url":"","msr_journal_url":"","msr_s2_pdf_url":"","msr_year":0,"msr_citation_count":0,"msr_influential_citations":0,"msr_reference_count":0,"msr_s2_match_confidence":0,"msr_microsoftintellectualproperty":true,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[],"msr_hide_image_in_river":0,"footnotes":""},"msr-research-highlight":[],"research-area":[13547],"msr-publication-type":[193716],"msr-publisher":[],"msr-focus-area":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-812314","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-systems-and-networking","msr-locale-en_us"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2022-1-1","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":"file","viewUrl":"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2022\/01\/MatchMaker_MLSys2022.pdf","id":"835198","title":"matchmaker_mlsys2022","label_id":"243109","label":0}],"msr_related_uploader":"","msr_citation_count":0,"msr_citation_count_updated":"","msr_s2_paper_id":"","msr_influential_citations":0,"msr_reference_count":0,"msr_arxiv_id":"","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[{"id":835198,"url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2022\/04\/MatchMaker_MLSys2022.pdf"},{"id":812317,"url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2022\/01\/MLSYS2022.pdf"}],"msr-author-ordering":[{"type":"edited_text","value":"Ankur Mallick","user_id":42441,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Ankur Mallick"},{"type":"edited_text","value":"Kevin Hsieh","user_id":39459,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Kevin Hsieh"},{"type":"edited_text","value":"Behnaz Arzani","user_id":37320,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Behnaz Arzani"},{"type":"text","value":"Gauri Joshi","user_id":0,"rest_url":false}],"msr_impact_theme":[],"msr_research_lab":[199565],"msr_event":[857133],"msr_group":[144899],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"inproceedings","related_content":[],"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/812314","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":5,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/812314\/revisions"}],"predecessor-version":[{"id":896931,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/812314\/revisions\/896931"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=812314"}],"wp:term":[{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=812314"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=812314"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=812314"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=812314"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=812314"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=812314"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=812314"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=812314"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=812314"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=812314"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=812314"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=812314"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}