Tailoring Mixup to Data for Calibration
Among all data augmentation techniques proposed so far, linear interpolation of training samples, also called Mixup, has found to be effective for a large panel of applications. Along with improved performance, Mixup is also a good technique for improving calibration and predictive uncertainty. Howe...
Gespeichert in:
Veröffentlicht in: | arXiv.org 2024-06 |
---|---|
Hauptverfasser: | , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
container_end_page | |
---|---|
container_issue | |
container_start_page | |
container_title | arXiv.org |
container_volume | |
creator | Bouniot, Quentin Mozharovskyi, Pavlo d'Alché-Buc, Florence |
description | Among all data augmentation techniques proposed so far, linear interpolation of training samples, also called Mixup, has found to be effective for a large panel of applications. Along with improved performance, Mixup is also a good technique for improving calibration and predictive uncertainty. However, mixing data carelessly can lead to manifold intrusion, i.e., conflicts between the synthetic labels assigned and the true label distributions, which can deteriorate calibration. In this work, we argue that the likelihood of manifold intrusion increases with the distance between data to mix. To this end, we propose to dynamically change the underlying distributions of interpolation coefficients depending on the similarity between samples to mix, and define a flexible framework to do so without losing in diversity. We provide extensive experiments for classification and regression tasks, showing that our proposed method improves performance and calibration of models, while being much more efficient. The code for our work is available at https://github.com/qbouniot/sim_kernel_mixup. |
format | Article |
fullrecord | <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_2885672514</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2885672514</sourcerecordid><originalsourceid>FETCH-proquest_journals_28856725143</originalsourceid><addsrcrecordid>eNpjYuA0MjY21LUwMTLiYOAtLs4yMDAwMjM3MjU15mRQD0nMzMkvysxLV_DNrCgtUCjJV3BJLElUSMsvUnBOzMlMKkosyczP42FgTUvMKU7lhdLcDMpuriHOHroFRfmFpanFJfFZ-aVFeUCpeCMLC1OQ8YYmxsSpAgBKvC90</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2885672514</pqid></control><display><type>article</type><title>Tailoring Mixup to Data for Calibration</title><source>Open Access: Freely Accessible Journals by multiple vendors</source><creator>Bouniot, Quentin ; Mozharovskyi, Pavlo ; d'Alché-Buc, Florence</creator><creatorcontrib>Bouniot, Quentin ; Mozharovskyi, Pavlo ; d'Alché-Buc, Florence</creatorcontrib><description>Among all data augmentation techniques proposed so far, linear interpolation of training samples, also called Mixup, has found to be effective for a large panel of applications. Along with improved performance, Mixup is also a good technique for improving calibration and predictive uncertainty. However, mixing data carelessly can lead to manifold intrusion, i.e., conflicts between the synthetic labels assigned and the true label distributions, which can deteriorate calibration. In this work, we argue that the likelihood of manifold intrusion increases with the distance between data to mix. To this end, we propose to dynamically change the underlying distributions of interpolation coefficients depending on the similarity between samples to mix, and define a flexible framework to do so without losing in diversity. We provide extensive experiments for classification and regression tasks, showing that our proposed method improves performance and calibration of models, while being much more efficient. The code for our work is available at https://github.com/qbouniot/sim_kernel_mixup.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Calibration ; Data augmentation ; Data points ; Deep learning ; Interpolation ; Machine learning ; Warping</subject><ispartof>arXiv.org, 2024-06</ispartof><rights>2024. This work is published under http://arxiv.org/licenses/nonexclusive-distrib/1.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>776,780</link.rule.ids></links><search><creatorcontrib>Bouniot, Quentin</creatorcontrib><creatorcontrib>Mozharovskyi, Pavlo</creatorcontrib><creatorcontrib>d'Alché-Buc, Florence</creatorcontrib><title>Tailoring Mixup to Data for Calibration</title><title>arXiv.org</title><description>Among all data augmentation techniques proposed so far, linear interpolation of training samples, also called Mixup, has found to be effective for a large panel of applications. Along with improved performance, Mixup is also a good technique for improving calibration and predictive uncertainty. However, mixing data carelessly can lead to manifold intrusion, i.e., conflicts between the synthetic labels assigned and the true label distributions, which can deteriorate calibration. In this work, we argue that the likelihood of manifold intrusion increases with the distance between data to mix. To this end, we propose to dynamically change the underlying distributions of interpolation coefficients depending on the similarity between samples to mix, and define a flexible framework to do so without losing in diversity. We provide extensive experiments for classification and regression tasks, showing that our proposed method improves performance and calibration of models, while being much more efficient. The code for our work is available at https://github.com/qbouniot/sim_kernel_mixup.</description><subject>Calibration</subject><subject>Data augmentation</subject><subject>Data points</subject><subject>Deep learning</subject><subject>Interpolation</subject><subject>Machine learning</subject><subject>Warping</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>ABUWG</sourceid><sourceid>AFKRA</sourceid><sourceid>AZQEC</sourceid><sourceid>BENPR</sourceid><sourceid>CCPQU</sourceid><sourceid>DWQXO</sourceid><recordid>eNpjYuA0MjY21LUwMTLiYOAtLs4yMDAwMjM3MjU15mRQD0nMzMkvysxLV_DNrCgtUCjJV3BJLElUSMsvUnBOzMlMKkosyczP42FgTUvMKU7lhdLcDMpuriHOHroFRfmFpanFJfFZ-aVFeUCpeCMLC1OQ8YYmxsSpAgBKvC90</recordid><startdate>20240611</startdate><enddate>20240611</enddate><creator>Bouniot, Quentin</creator><creator>Mozharovskyi, Pavlo</creator><creator>d'Alché-Buc, Florence</creator><general>Cornell University Library, arXiv.org</general><scope>8FE</scope><scope>8FG</scope><scope>ABJCF</scope><scope>ABUWG</scope><scope>AFKRA</scope><scope>AZQEC</scope><scope>BENPR</scope><scope>BGLVJ</scope><scope>CCPQU</scope><scope>DWQXO</scope><scope>HCIFZ</scope><scope>L6V</scope><scope>M7S</scope><scope>PIMPY</scope><scope>PQEST</scope><scope>PQQKQ</scope><scope>PQUKI</scope><scope>PRINS</scope><scope>PTHSS</scope></search><sort><creationdate>20240611</creationdate><title>Tailoring Mixup to Data for Calibration</title><author>Bouniot, Quentin ; Mozharovskyi, Pavlo ; d'Alché-Buc, Florence</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_28856725143</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Calibration</topic><topic>Data augmentation</topic><topic>Data points</topic><topic>Deep learning</topic><topic>Interpolation</topic><topic>Machine learning</topic><topic>Warping</topic><toplevel>online_resources</toplevel><creatorcontrib>Bouniot, Quentin</creatorcontrib><creatorcontrib>Mozharovskyi, Pavlo</creatorcontrib><creatorcontrib>d'Alché-Buc, Florence</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science & Engineering Collection</collection><collection>ProQuest Central (Alumni)</collection><collection>ProQuest Central</collection><collection>ProQuest Central Essentials</collection><collection>ProQuest Central</collection><collection>Technology Collection</collection><collection>ProQuest One Community College</collection><collection>ProQuest Central</collection><collection>SciTech Premium Collection</collection><collection>ProQuest Engineering Collection</collection><collection>Engineering Database</collection><collection>Publicly Available Content Database</collection><collection>ProQuest One Academic Eastern Edition (DO NOT USE)</collection><collection>ProQuest One Academic</collection><collection>ProQuest One Academic UKI Edition</collection><collection>ProQuest Central China</collection><collection>Engineering collection</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Bouniot, Quentin</au><au>Mozharovskyi, Pavlo</au><au>d'Alché-Buc, Florence</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Tailoring Mixup to Data for Calibration</atitle><jtitle>arXiv.org</jtitle><date>2024-06-11</date><risdate>2024</risdate><eissn>2331-8422</eissn><abstract>Among all data augmentation techniques proposed so far, linear interpolation of training samples, also called Mixup, has found to be effective for a large panel of applications. Along with improved performance, Mixup is also a good technique for improving calibration and predictive uncertainty. However, mixing data carelessly can lead to manifold intrusion, i.e., conflicts between the synthetic labels assigned and the true label distributions, which can deteriorate calibration. In this work, we argue that the likelihood of manifold intrusion increases with the distance between data to mix. To this end, we propose to dynamically change the underlying distributions of interpolation coefficients depending on the similarity between samples to mix, and define a flexible framework to do so without losing in diversity. We provide extensive experiments for classification and regression tasks, showing that our proposed method improves performance and calibration of models, while being much more efficient. The code for our work is available at https://github.com/qbouniot/sim_kernel_mixup.</abstract><cop>Ithaca</cop><pub>Cornell University Library, arXiv.org</pub><oa>free_for_read</oa></addata></record> |
fulltext | fulltext |
identifier | EISSN: 2331-8422 |
ispartof | arXiv.org, 2024-06 |
issn | 2331-8422 |
language | eng |
recordid | cdi_proquest_journals_2885672514 |
source | Open Access: Freely Accessible Journals by multiple vendors |
subjects | Calibration Data augmentation Data points Deep learning Interpolation Machine learning Warping |
title | Tailoring Mixup to Data for Calibration |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-24T19%3A10%3A19IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=document&rft.atitle=Tailoring%20Mixup%20to%20Data%20for%20Calibration&rft.jtitle=arXiv.org&rft.au=Bouniot,%20Quentin&rft.date=2024-06-11&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E2885672514%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2885672514&rft_id=info:pmid/&rfr_iscdi=true |