{"id":534,"date":"2018-04-07T12:55:34","date_gmt":"2018-04-07T12:55:34","guid":{"rendered":"http:\/\/web.eecs.umich.edu\/~girasole\/?page_id=534"},"modified":"2018-04-07T12:55:34","modified_gmt":"2018-04-07T12:55:34","slug":"active-learning","status":"publish","type":"page","link":"https:\/\/web.eecs.umich.edu\/~girasole\/?page_id=534","title":{"rendered":"Active Learning"},"content":{"rendered":"

In many applications, it’s possible to take measurements repeatedly to use for inference. Examples include genetic experiments, environmental sensing, and crowdsourced image tasks. The problem of how to design sequential measurements for machine learning inference is called active learning. We can and should exploit expert knowledge about the signal of interest. However, if we trust a model too much, we may miss a true signal. We have studied active learning algorithms for image clustering\/classification and spatial environmental sampling.<\/p>\n

\n\n\t\t1399621<\/span>\n\t\t<\/span>\n\t\tE36HLPSJ<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n <\/span>\n\t\t1<\/span>\n\t\tchicago-author-date<\/span>\n\t\t50<\/span>\n\t\tdate<\/span>\n\t\tdesc<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t1<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n <\/span>\n 534<\/span>\n\t\thttps:\/\/web.eecs.umich.edu\/~girasole\/wp-content\/plugins\/zotpress\/<\/span>\n\n\t\t
\n\t\t\t
\n\t\t\t\t%7B%22status%22%3A%22success%22%2C%22updateneeded%22%3Afalse%2C%22instance%22%3Afalse%2C%22meta%22%3A%7B%22request_last%22%3A0%2C%22request_next%22%3A0%2C%22used_cache%22%3Atrue%7D%2C%22data%22%3A%5B%7B%22key%22%3A%22DTH4F5EN%22%2C%22library%22%3A%7B%22id%22%3A1399621%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Lipor%20et%20al.%22%2C%22parsedDate%22%3A%222017-10%22%2C%22numChildren%22%3A1%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%201.35%3B%20padding-left%3A%201em%3B%20text-indent%3A-1em%3B%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%26gt%3BLipor%2C%20J.%2C%20B.%20P.%20Wong%2C%20D.%20Scavia%2C%20B.%20Kerkez%2C%20and%20L.%20Balzano.%202017.%20%26%23x201C%3BDistance-Penalized%20Active%20Learning%20Using%20Quantile%20Search.%26%23x201D%3B%20%26lt%3Bi%26gt%3BIEEE%20Transactions%20on%20Signal%20Processing%26lt%3B%5C%2Fi%26gt%3B%2065%20%2820%29%3A%205453%26%23x2013%3B65.%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FTSP.2017.2731323%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FTSP.2017.2731323%26lt%3B%5C%2Fa%26gt%3B.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Distance-Penalized%20Active%20Learning%20Using%20Quantile%20Search%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22J.%22%2C%22lastName%22%3A%22Lipor%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22B.%20P.%22%2C%22lastName%22%3A%22Wong%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22D.%22%2C%22lastName%22%3A%22Scavia%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22B.%22%2C%22lastName%22%3A%22Kerkez%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22L.%22%2C%22lastName%22%3A%22Balzano%22%7D%5D%2C%22abstractNote%22%3A%22Adaptive%20sampling%20theory%20has%20shown%20that%2C%20with%20proper%20assumptions%20on%20the%20signal%20class%2C%20algorithms%20exist%20to%20reconstruct%20a%20signal%20in%20%24mathbb%20R%5Ed%24%20with%20an%20optimal%20number%20of%20samples.%20We%20generalize%20this%20problem%20to%20the%20case%20of%20spatial%20signals%2C%20where%20the%20sampling%20cost%20is%20a%20function%20of%20both%20the%20number%20of%20samples%20taken%20and%20the%20distance%20traveled%20during%20estimation.%20This%20is%20motivated%20by%20our%20work%20studying%20regions%20of%20low%20oxygen%20concentration%20in%20the%20Great%20Lakes.%20We%20show%20that%20for%20one-dimensional%20threshold%20classifiers%2C%20a%20tradeoff%20between%20the%20number%20of%20samples%20taken%20and%20distance%20traveled%20can%20be%20achieved%20using%20a%20generalization%20of%20binary%20search%2C%20which%20we%20refer%20to%20as%20quantile%20search.%20We%20characterize%20both%20the%20estimation%20error%20after%20a%20fixed%20number%20of%20samples%20and%20the%20distance%20traveled%20in%20the%20noiseless%20case%2C%20as%20well%20as%20the%20estimation%20error%20in%20the%20case%20of%20noisy%20measurements.%20We%20illustrate%20our%20results%20in%20both%20simulations%20and%20experiments%20and%20show%20that%20our%20method%20outperforms%20existing%20algorithms%20in%20a%20large%20range%20of%20sampling%20scenarios.%22%2C%22date%22%3A%22October%202017%22%2C%22section%22%3A%22%22%2C%22partNumber%22%3A%22%22%2C%22partTitle%22%3A%22%22%2C%22DOI%22%3A%2210.1109%5C%2FTSP.2017.2731323%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22%22%2C%22PMID%22%3A%22%22%2C%22PMCID%22%3A%22%22%2C%22ISSN%22%3A%221053-587X%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22427SEM27%22%2C%22DZFDBB6V%22%2C%22E36HLPSJ%22%2C%22UIWU664R%22%5D%2C%22dateModified%22%3A%222018-02-16T18%3A44%3A47Z%22%7D%7D%2C%7B%22key%22%3A%2278Q9AHXX%22%2C%22library%22%3A%7B%22id%22%3A1399621%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Lipor%20and%20Balzano%22%2C%22parsedDate%22%3A%222015-12%22%2C%22numChildren%22%3A2%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%201.35%3B%20padding-left%3A%201em%3B%20text-indent%3A-1em%3B%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%26gt%3BLipor%2C%20J.%2C%20and%20L.%20Balzano.%202015.%20%26%23x201C%3BMargin-Based%20Active%20Subspace%20Clustering.%26%23x201D%3B%20%26lt%3Bi%26gt%3B2015%20IEEE%206th%20International%20Workshop%20on%20Computational%20Advances%20in%20Multi-Sensor%20Adaptive%20Processing%20%28CAMSAP%29%26lt%3B%5C%2Fi%26gt%3B%2C%20December%2C%20377%26%23x2013%3B80.%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FCAMSAP.2015.7383815%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FCAMSAP.2015.7383815%26lt%3B%5C%2Fa%26gt%3B.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22conferencePaper%22%2C%22title%22%3A%22Margin-based%20active%20subspace%20clustering%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22J.%22%2C%22lastName%22%3A%22Lipor%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22L.%22%2C%22lastName%22%3A%22Balzano%22%7D%5D%2C%22abstractNote%22%3A%22Subspace%20clustering%20has%20typically%20been%20approached%20as%20an%20unsupervised%20machine%20learning%20problem.%20However%20in%20several%20applications%20where%20the%20union%20of%20subspaces%20model%20is%20useful%2C%20it%20is%20also%20reasonable%20to%20assume%20you%20have%20access%20to%20a%20small%20number%20of%20labels.%20In%20this%20paper%20we%20investigate%20the%20benefit%20labeled%20data%20brings%20to%20the%20subspace%20clustering%20problem.%20We%20focus%20on%20incorporating%20labels%20into%20the%20k-subspaces%20algorithm%2C%20a%20simple%20and%20computationally%20efficient%20alternating%20estimation%20algorithm.%20We%20find%20that%20even%20a%20very%20small%20number%20of%20randomly%20selected%20labels%20can%20greatly%20improve%20accuracy%20over%20the%20unsupervised%20approach.%20We%20demonstrate%20that%20with%20enough%20labels%2C%20we%20get%20a%20significant%20improvement%20by%20using%20actively%20selected%20labels%20chosen%20for%20points%20that%20are%20nearly%20equidistant%20to%20more%20than%20one%20estimated%20subspace.%20We%20show%20this%20improvement%20on%20simulated%20data%20and%20face%20images.%22%2C%22proceedingsTitle%22%3A%222015%20IEEE%206th%20International%20Workshop%20on%20Computational%20Advances%20in%20Multi-Sensor%20Adaptive%20Processing%20%28CAMSAP%29%22%2C%22conferenceName%22%3A%222015%20IEEE%206th%20International%20Workshop%20on%20Computational%20Advances%20in%20Multi-Sensor%20Adaptive%20Processing%20%28CAMSAP%29%22%2C%22date%22%3A%22December%202015%22%2C%22eventPlace%22%3A%22%22%2C%22DOI%22%3A%2210.1109%5C%2FCAMSAP.2015.7383815%22%2C%22ISBN%22%3A%22%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22ieeexplore.ieee.org%5C%2Fxpl%5C%2FarticleDetails.jsp%3Farnumber%3D7383815%22%2C%22ISSN%22%3A%22%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%226JKB3X7P%22%2C%22DZFDBB6V%22%2C%22E36HLPSJ%22%2C%22ZA8QMDGD%22%5D%2C%22dateModified%22%3A%222016-06-13T15%3A59%3A42Z%22%7D%7D%2C%7B%22key%22%3A%22UUTZBDMJ%22%2C%22library%22%3A%7B%22id%22%3A1399621%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Lipor%20et%20al.%22%2C%22parsedDate%22%3A%222015-09%22%2C%22numChildren%22%3A1%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%201.35%3B%20padding-left%3A%201em%3B%20text-indent%3A-1em%3B%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%26gt%3BLipor%2C%20J.%2C%20L.%20Balzano%2C%20B.%20Kerkez%2C%20and%20D.%20Scavia.%202015.%20%26%23x201C%3BQuantile%20Search%3A%20A%20Distance-Penalized%20Active%20Learning%20Algorithm%20for%20Spatial%20Sampling.%26%23x201D%3B%20%26lt%3Bi%26gt%3B2015%2053rd%20Annual%20Allerton%20Conference%20on%20Communication%2C%20Control%2C%20and%20Computing%20%28Allerton%29%26lt%3B%5C%2Fi%26gt%3B%2C%20September%2C%201241%26%23x2013%3B48.%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FALLERTON.2015.7447150%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FALLERTON.2015.7447150%26lt%3B%5C%2Fa%26gt%3B.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22conferencePaper%22%2C%22title%22%3A%22Quantile%20search%3A%20A%20distance-penalized%20active%20learning%20algorithm%20for%20spatial%20sampling%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22J.%22%2C%22lastName%22%3A%22Lipor%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22L.%22%2C%22lastName%22%3A%22Balzano%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22B.%22%2C%22lastName%22%3A%22Kerkez%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22D.%22%2C%22lastName%22%3A%22Scavia%22%7D%5D%2C%22abstractNote%22%3A%22Adaptive%20sampling%20theory%20has%20shown%20that%2C%20with%20proper%20assumptions%20on%20the%20signal%20class%2C%20algorithms%20exist%20to%20reconstruct%20a%20signal%20in%20%3F%3F%3Fd%20with%20an%20optimal%20number%20of%20samples.%20We%20generalize%20this%20problem%20to%20when%20the%20cost%20of%20sampling%20is%20not%20only%20the%20number%20of%20samples%20but%20also%20the%20distance%20traveled%20between%20samples.%20This%20is%20motivated%20by%20our%20work%20studying%20regions%20of%20low%20oxygen%20concentration%20in%20the%20Great%20Lakes.%20We%20show%20that%20for%20one-dimensional%20threshold%20classifiers%2C%20a%20tradeoff%20between%20number%20of%20samples%20and%20distance%20traveled%20can%20be%20achieved%20using%20a%20generalization%20of%20binary%20search%2C%20which%20we%20refer%20to%20as%20quantile%20search.%20We%20derive%20the%20expected%20total%20sampling%20time%20for%20noiseless%20measurements%20and%20the%20expected%20number%20of%20samples%20for%20an%20extension%20to%20the%20noisy%20case.%20We%20illustrate%20our%20results%20in%20simulations%20relevant%20to%20our%20sampling%20application.%22%2C%22proceedingsTitle%22%3A%222015%2053rd%20Annual%20Allerton%20Conference%20on%20Communication%2C%20Control%2C%20and%20Computing%20%28Allerton%29%22%2C%22conferenceName%22%3A%222015%2053rd%20Annual%20Allerton%20Conference%20on%20Communication%2C%20Control%2C%20and%20Computing%20%28Allerton%29%22%2C%22date%22%3A%22Sept%202015%22%2C%22eventPlace%22%3A%22%22%2C%22DOI%22%3A%2210.1109%5C%2FALLERTON.2015.7447150%22%2C%22ISBN%22%3A%22%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Fieeexplore.ieee.org%5C%2Fxpl%5C%2FarticleDetails.jsp%3Farnumber%3D7447150%22%2C%22ISSN%22%3A%22%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22DZFDBB6V%22%2C%22E36HLPSJ%22%2C%22UIWU664R%22%2C%22ZA8QMDGD%22%5D%2C%22dateModified%22%3A%222016-06-13T16%3A02%3A06Z%22%7D%7D%5D%7D<\/span>\n\n\t\t\t\t
\n
\n
Lipor, J., B. P. Wong, D. Scavia, B. Kerkez, and L. Balzano. 2017. \u201cDistance-Penalized Active Learning Using Quantile Search.\u201d IEEE Transactions on Signal Processing<\/i> 65 (20): 5453\u201365. https:\/\/doi.org\/10.1109\/TSP.2017.2731323<\/a>.<\/div>\n<\/div>\n\t\t\t\t<\/div>\t\t\t\t
\n
\n
Lipor, J., and L. Balzano. 2015. \u201cMargin-Based Active Subspace Clustering.\u201d 2015 IEEE 6th International Workshop on Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP)<\/i>, December, 377\u201380. https:\/\/doi.org\/10.1109\/CAMSAP.2015.7383815<\/a>.<\/div>\n<\/div>\n\t\t\t\t<\/div>\t\t\t\t
\n
\n
Lipor, J., L. Balzano, B. Kerkez, and D. Scavia. 2015. \u201cQuantile Search: A Distance-Penalized Active Learning Algorithm for Spatial Sampling.\u201d 2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)<\/i>, September, 1241\u201348. https:\/\/doi.org\/10.1109\/ALLERTON.2015.7447150<\/a>.<\/div>\n<\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\n\n","protected":false},"excerpt":{"rendered":"

In many applications, it’s possible to take measurements repeatedly to use for inference. Examples include genetic experiments, environmental sensing, and crowdsourced image tasks. The problem of how to design sequential measurements for machine learning inference is called active learning. We can and should exploit expert knowledge about the signal of interest. However, if we trust […]<\/p>\n","protected":false},"author":3,"featured_media":0,"parent":182,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":[],"_links":{"self":[{"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/pages\/534"}],"collection":[{"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=534"}],"version-history":[{"count":1,"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/pages\/534\/revisions"}],"predecessor-version":[{"id":535,"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/pages\/534\/revisions\/535"}],"up":[{"embeddable":true,"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/pages\/182"}],"wp:attachment":[{"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=534"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}