{"id":183492,"date":"2006-07-10T00:00:00","date_gmt":"2009-10-31T12:46:00","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/msr-research-item\/graph-cuts-without-eigenvectors\/"},"modified":"2016-09-09T09:51:08","modified_gmt":"2016-09-09T16:51:08","slug":"graph-cuts-without-eigenvectors","status":"publish","type":"msr-video","link":"https:\/\/www.microsoft.com\/en-us\/research\/video\/graph-cuts-without-eigenvectors\/","title":{"rendered":"Graph Cuts without Eigenvectors"},"content":{"rendered":"<div class=\"asset-content\">\n<p>Graph clustering&#8212;clustering the nodes of a graph&#8212;is a fundamental problem arising in many machine learning and data mining applications.  The last few years have seen a surge of interest in spectral clustering methods, which use eigenvectors of the graph adjacency matrix (or a matrix derived from the adjacency matrix) to optimize one of several graph cut objective functions.  These methods are powerful and theoretically well-motivated; however, computing eigenvectors of a large matrix is computationally expensive and requires significant memory overhead.<\/p>\n<p>I will describe my recent research on optimizing spectral clustering objectives without costly eigenvector computation.  After deriving a mathematical connection between spectral clustering objectives and weighted kernel k-means, I will propose a simple iterative algorithm for optimizing various graph cut objectives, and then embed this into a multilevel algorithm.  This resulting algorithm outperforms state-of-the-art spectral clustering algorithms in terms of quality (graph cut value), speed (up to 2000 times faster on graphs under a million nodes), and memory usage (requiring memory on the order of the input graph).  I will also discuss applications of this algorithm to image segmentation, protein network clustering, and social network analysis.<\/p>\n<p>This is joint work with Inderjit Dhillon and Yuqiang Guan.<\/p>\n<\/div>\n<p><!-- .asset-content --><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Graph clustering&#8212;clustering the nodes of a graph&#8212;is a fundamental problem arising in many machine learning and data mining applications. The last few years have seen a surge of interest in spectral clustering methods, which use eigenvectors of the graph adjacency matrix (or a matrix derived from the adjacency matrix) to optimize one of several graph [&hellip;]<\/p>\n","protected":false},"featured_media":195057,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr_hide_image_in_river":0,"footnotes":""},"research-area":[],"msr-video-type":[],"msr-locale":[268875],"msr-post-option":[],"msr-session-type":[],"msr-impact-theme":[],"msr-pillar":[],"msr-episode":[],"msr-research-theme":[],"class_list":["post-183492","msr-video","type-msr-video","status-publish","has-post-thumbnail","hentry","msr-locale-en_us"],"msr_download_urls":"","msr_external_url":"https:\/\/youtu.be\/kksLJ2D5wdU","msr_secondary_video_url":"","msr_video_file":"","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/183492","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":0,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/183492\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/195057"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=183492"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=183492"},{"taxonomy":"msr-video-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video-type?post=183492"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=183492"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=183492"},{"taxonomy":"msr-session-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-session-type?post=183492"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=183492"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=183492"},{"taxonomy":"msr-episode","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-episode?post=183492"},{"taxonomy":"msr-research-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-theme?post=183492"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}