Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations

Neural network classifiers can largely rely on simple spurious features, such as backgrounds, to make predictions. However, even in these cases, we show that they still often learn core features associated with the desired attributes of the data, contrary to recent findings. Inspired by this insight...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Kirichenko, Polina, Izmailov, Pavel, Wilson, Andrew Gordon
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Kirichenko, Polina
Izmailov, Pavel
Wilson, Andrew Gordon
description Neural network classifiers can largely rely on simple spurious features, such as backgrounds, to make predictions. However, even in these cases, we show that they still often learn core features associated with the desired attributes of the data, contrary to recent findings. Inspired by this insight, we demonstrate that simple last layer retraining can match or outperform state-of-the-art approaches on spurious correlation benchmarks, but with profoundly lower complexity and computational expenses. Moreover, we show that last layer retraining on large ImageNet-trained models can also significantly reduce reliance on background and texture information, improving robustness to covariate shift, after only minutes of training on a single GPU.
doi_str_mv 10.48550/arxiv.2204.02937
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2204_02937</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2204_02937</sourcerecordid><originalsourceid>FETCH-LOGICAL-a677-294f26f61d74cf54806de538d8215dadd11b60251c45dec44c7f1020e2d4e3353</originalsourceid><addsrcrecordid>eNotz81KAzEYheFsXEjrBbgyNzBj_jNdlsGqEBDs7Ic0-SKBmpQkI-3dq9XVWbxw4EHonpJeDFKSR1vO8atnjIiesA3Xt8gYWxs29gIFv0M3FRtTTB84VrxfQoguQmo45J-aD0ttCWrFLeP9aSkxLxWPuRQ42hZzqmt0E-yxwt3_rtC0e5rGl868Pb-OW9NZpXXHNiIwFRT1WrggxUCUB8kHPzAqvfWe0oMiTFInpAcnhNOBEkaAeQGcS75CD3-3V858KvHTlsv8y5qvLP4N2YpH4w</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations</title><source>arXiv.org</source><creator>Kirichenko, Polina ; Izmailov, Pavel ; Wilson, Andrew Gordon</creator><creatorcontrib>Kirichenko, Polina ; Izmailov, Pavel ; Wilson, Andrew Gordon</creatorcontrib><description>Neural network classifiers can largely rely on simple spurious features, such as backgrounds, to make predictions. However, even in these cases, we show that they still often learn core features associated with the desired attributes of the data, contrary to recent findings. Inspired by this insight, we demonstrate that simple last layer retraining can match or outperform state-of-the-art approaches on spurious correlation benchmarks, but with profoundly lower complexity and computational expenses. Moreover, we show that last layer retraining on large ImageNet-trained models can also significantly reduce reliance on background and texture information, improving robustness to covariate shift, after only minutes of training on a single GPU.</description><identifier>DOI: 10.48550/arxiv.2204.02937</identifier><language>eng</language><subject>Computer Science - Computer Vision and Pattern Recognition ; Computer Science - Learning ; Statistics - Machine Learning</subject><creationdate>2022-04</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,780,885</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2204.02937$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2204.02937$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Kirichenko, Polina</creatorcontrib><creatorcontrib>Izmailov, Pavel</creatorcontrib><creatorcontrib>Wilson, Andrew Gordon</creatorcontrib><title>Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations</title><description>Neural network classifiers can largely rely on simple spurious features, such as backgrounds, to make predictions. However, even in these cases, we show that they still often learn core features associated with the desired attributes of the data, contrary to recent findings. Inspired by this insight, we demonstrate that simple last layer retraining can match or outperform state-of-the-art approaches on spurious correlation benchmarks, but with profoundly lower complexity and computational expenses. Moreover, we show that last layer retraining on large ImageNet-trained models can also significantly reduce reliance on background and texture information, improving robustness to covariate shift, after only minutes of training on a single GPU.</description><subject>Computer Science - Computer Vision and Pattern Recognition</subject><subject>Computer Science - Learning</subject><subject>Statistics - Machine Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2022</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotz81KAzEYheFsXEjrBbgyNzBj_jNdlsGqEBDs7Ic0-SKBmpQkI-3dq9XVWbxw4EHonpJeDFKSR1vO8atnjIiesA3Xt8gYWxs29gIFv0M3FRtTTB84VrxfQoguQmo45J-aD0ttCWrFLeP9aSkxLxWPuRQ42hZzqmt0E-yxwt3_rtC0e5rGl868Pb-OW9NZpXXHNiIwFRT1WrggxUCUB8kHPzAqvfWe0oMiTFInpAcnhNOBEkaAeQGcS75CD3-3V858KvHTlsv8y5qvLP4N2YpH4w</recordid><startdate>20220406</startdate><enddate>20220406</enddate><creator>Kirichenko, Polina</creator><creator>Izmailov, Pavel</creator><creator>Wilson, Andrew Gordon</creator><scope>AKY</scope><scope>EPD</scope><scope>GOX</scope></search><sort><creationdate>20220406</creationdate><title>Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations</title><author>Kirichenko, Polina ; Izmailov, Pavel ; Wilson, Andrew Gordon</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a677-294f26f61d74cf54806de538d8215dadd11b60251c45dec44c7f1020e2d4e3353</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2022</creationdate><topic>Computer Science - Computer Vision and Pattern Recognition</topic><topic>Computer Science - Learning</topic><topic>Statistics - Machine Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Kirichenko, Polina</creatorcontrib><creatorcontrib>Izmailov, Pavel</creatorcontrib><creatorcontrib>Wilson, Andrew Gordon</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv Statistics</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Kirichenko, Polina</au><au>Izmailov, Pavel</au><au>Wilson, Andrew Gordon</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations</atitle><date>2022-04-06</date><risdate>2022</risdate><abstract>Neural network classifiers can largely rely on simple spurious features, such as backgrounds, to make predictions. However, even in these cases, we show that they still often learn core features associated with the desired attributes of the data, contrary to recent findings. Inspired by this insight, we demonstrate that simple last layer retraining can match or outperform state-of-the-art approaches on spurious correlation benchmarks, but with profoundly lower complexity and computational expenses. Moreover, we show that last layer retraining on large ImageNet-trained models can also significantly reduce reliance on background and texture information, improving robustness to covariate shift, after only minutes of training on a single GPU.</abstract><doi>10.48550/arxiv.2204.02937</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2204.02937
ispartof
issn
language eng
recordid cdi_arxiv_primary_2204_02937
source arXiv.org
subjects Computer Science - Computer Vision and Pattern Recognition
Computer Science - Learning
Statistics - Machine Learning
title Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-23T13%3A12%3A02IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Last%20Layer%20Re-Training%20is%20Sufficient%20for%20Robustness%20to%20Spurious%20Correlations&rft.au=Kirichenko,%20Polina&rft.date=2022-04-06&rft_id=info:doi/10.48550/arxiv.2204.02937&rft_dat=%3Carxiv_GOX%3E2204_02937%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true