{"id":490424,"date":"2018-06-09T12:14:07","date_gmt":"2018-06-09T19:14:07","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-research-item&p=490424"},"modified":"2018-11-05T15:16:28","modified_gmt":"2018-11-05T23:16:28","slug":"blas-flash-efficient-alternative-large-scale-ml-training-inference","status":"publish","type":"msr-research-item","link":"https:\/\/www.microsoft.com\/en-us\/research\/publication\/blas-flash-efficient-alternative-large-scale-ml-training-inference\/","title":{"rendered":"BLAS-on-flash : An Efficient Alternative for Large Scale ML Training and Inference?"},"content":{"rendered":"

Code Release (opens in new tab)<\/span><\/a><\/p>\n

Many large-scale machine learning training and inference tasks are memory-bound rather than compute-bound. That is, on large data sets, the working set of these algorithms does not fit in memory for jobs that could run overnight on a few multi-core processors. This often forces an expensive redesign of the algorithm for distributed platforms such as parameter servers and Spark.<\/p>\n

We propose an inexpensive and efficient alternative based on the observation that many ML tasks admit algorithms that can be programmed with linear algebra subroutines. A library that supports BLAS and sparseBLAS interface on large SSD-resident matrices can enable multi-threaded code to scale to industrial scale datasets on a single workstation.<\/p>\n

We demonstrate that not only can such a library provide near in-memory performance for BLAS, but can also be used to write implementations of complex algorithms such as eigensolvers that outperform in-memory (ARPACK) and distributed (Spark) counterparts.<\/p>\n

Existing multi-threaded in-memory code can link to our library with minor changes and scale to hundreds of gigabytes of training or inference data at near in-memory processing speeds. We demonstrate this with two industrial scale use cases arising in ranking and relevance pipelines: training large scale topic models and inference for extreme multi-label learning.<\/p>\n

This suggests that our approach could be an efficient alternative to expensive distributed big-data systems for scaling up structurally complex machine learning tasks.<\/p>\n","protected":false},"excerpt":{"rendered":"

Code Release Many large-scale machine learning training and inference tasks are memory-bound rather than compute-bound. That is, on large data sets, the working set of these algorithms does not fit in memory for jobs that could run overnight on a few multi-core processors. This often forces an expensive redesign of the algorithm for distributed platforms […]<\/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":[13561,13547],"msr-publication-type":[193716],"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-490424","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-algorithms","msr-research-area-systems-and-networking","msr-locale-en_us"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2019-2-1","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":"497450","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"file","viewUrl":"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/06\/main-5b59716a59e37.pdf","id":"497450","title":"main-5b59716a59e37","label_id":"243109","label":0}],"msr_related_uploader":"","msr_attachments":[{"id":544941,"url":"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/10\/nsdi19_final.pdf"},{"id":497450,"url":"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/07\/main-5b59716a59e37.pdf"}],"msr-author-ordering":[{"type":"edited_text","value":"Harsha Simhadri","user_id":36146,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Harsha Simhadri"},{"type":"text","value":"Suhas Jayaram Subramanya","user_id":0,"rest_url":false},{"type":"text","value":"Srajan Garg","user_id":0,"rest_url":false},{"type":"text","value":"Anil Kag","user_id":0,"rest_url":false},{"type":"text","value":"B. Venkatesh","user_id":0,"rest_url":false}],"msr_impact_theme":[],"msr_research_lab":[199562],"msr_event":[],"msr_group":[],"msr_project":[],"publication":[],"video":[],"download":[],"msr_publication_type":"inproceedings","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/490424"}],"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\/490424\/revisions"}],"predecessor-version":[{"id":544968,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/490424\/revisions\/544968"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=490424"}],"wp:term":[{"taxonomy":"msr-content-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-content-type?post=490424"},{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=490424"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=490424"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=490424"},{"taxonomy":"msr-product-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-product-type?post=490424"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=490424"},{"taxonomy":"msr-platform","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-platform?post=490424"},{"taxonomy":"msr-download-source","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-download-source?post=490424"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=490424"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=490424"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=490424"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=490424"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=490424"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=490424"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}