Two Sides of Miscalibration: Identifying Over and Under-Confidence Prediction for Network Calibration

Proper confidence calibration of deep neural networks is essential for reliable predictions in safety-critical tasks. Miscalibration can lead to model over-confidence and/or under-confidence; i.e., the model's confidence in its prediction can be greater or less than the model's accuracy. R...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:arXiv.org 2023-08
Hauptverfasser: Ao, Shuang, Rueger, Stefan, Siddharthan, Advaith
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 Ao, Shuang
Rueger, Stefan
Siddharthan, Advaith
description Proper confidence calibration of deep neural networks is essential for reliable predictions in safety-critical tasks. Miscalibration can lead to model over-confidence and/or under-confidence; i.e., the model's confidence in its prediction can be greater or less than the model's accuracy. Recent studies have highlighted the over-confidence issue by introducing calibration techniques and demonstrated success on various tasks. However, miscalibration through under-confidence has not yet to receive much attention. In this paper, we address the necessity of paying attention to the under-confidence issue. We first introduce a novel metric, a miscalibration score, to identify the overall and class-wise calibration status, including being over or under-confident. Our proposed metric reveals the pitfalls of existing calibration techniques, where they often overly calibrate the model and worsen under-confident predictions. Then we utilize the class-wise miscalibration score as a proxy to design a calibration technique that can tackle both over and under-confidence. We report extensive experiments that show our proposed methods substantially outperforming existing calibration techniques. We also validate our proposed calibration technique on an automatic failure detection task with a risk-coverage curve, reporting that our methods improve failure detection as well as trustworthiness of the model. The code are available at \url{https://github.com/AoShuang92/miscalibration_TS}.
format Article
fullrecord <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_2847573792</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2847573792</sourcerecordid><originalsourceid>FETCH-proquest_journals_28475737923</originalsourceid><addsrcrecordid>eNqNjE8LgjAYh0cQJOV3eKGzYJumdZWiDv2B7CzLvYuZbLVp0rfPIOja6Xd4nuc3IB5lbBakEaUj4jtXhWFI5wmNY-YRzDsDJyXQgZGwU67ktbpY3iijl7AVqBslX0pf4fBEC1wLOGuBNsiMln2mS4SjRaHKTwHSWNhj0xl7g-z3NCFDyWuH_nfHZLpe5dkmuFvzaNE1RWVaq3tU0DRK4oQlC8r-s95FyUa0</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2847573792</pqid></control><display><type>article</type><title>Two Sides of Miscalibration: Identifying Over and Under-Confidence Prediction for Network Calibration</title><source>Free E- Journals</source><creator>Ao, Shuang ; Rueger, Stefan ; Siddharthan, Advaith</creator><creatorcontrib>Ao, Shuang ; Rueger, Stefan ; Siddharthan, Advaith</creatorcontrib><description>Proper confidence calibration of deep neural networks is essential for reliable predictions in safety-critical tasks. Miscalibration can lead to model over-confidence and/or under-confidence; i.e., the model's confidence in its prediction can be greater or less than the model's accuracy. Recent studies have highlighted the over-confidence issue by introducing calibration techniques and demonstrated success on various tasks. However, miscalibration through under-confidence has not yet to receive much attention. In this paper, we address the necessity of paying attention to the under-confidence issue. We first introduce a novel metric, a miscalibration score, to identify the overall and class-wise calibration status, including being over or under-confident. Our proposed metric reveals the pitfalls of existing calibration techniques, where they often overly calibrate the model and worsen under-confident predictions. Then we utilize the class-wise miscalibration score as a proxy to design a calibration technique that can tackle both over and under-confidence. We report extensive experiments that show our proposed methods substantially outperforming existing calibration techniques. We also validate our proposed calibration technique on an automatic failure detection task with a risk-coverage curve, reporting that our methods improve failure detection as well as trustworthiness of the model. The code are available at \url{https://github.com/AoShuang92/miscalibration_TS}.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Artificial neural networks ; Calibration ; Confidence ; Failure detection ; Model accuracy ; Safety critical</subject><ispartof>arXiv.org, 2023-08</ispartof><rights>2023. 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>Ao, Shuang</creatorcontrib><creatorcontrib>Rueger, Stefan</creatorcontrib><creatorcontrib>Siddharthan, Advaith</creatorcontrib><title>Two Sides of Miscalibration: Identifying Over and Under-Confidence Prediction for Network Calibration</title><title>arXiv.org</title><description>Proper confidence calibration of deep neural networks is essential for reliable predictions in safety-critical tasks. Miscalibration can lead to model over-confidence and/or under-confidence; i.e., the model's confidence in its prediction can be greater or less than the model's accuracy. Recent studies have highlighted the over-confidence issue by introducing calibration techniques and demonstrated success on various tasks. However, miscalibration through under-confidence has not yet to receive much attention. In this paper, we address the necessity of paying attention to the under-confidence issue. We first introduce a novel metric, a miscalibration score, to identify the overall and class-wise calibration status, including being over or under-confident. Our proposed metric reveals the pitfalls of existing calibration techniques, where they often overly calibrate the model and worsen under-confident predictions. Then we utilize the class-wise miscalibration score as a proxy to design a calibration technique that can tackle both over and under-confidence. We report extensive experiments that show our proposed methods substantially outperforming existing calibration techniques. We also validate our proposed calibration technique on an automatic failure detection task with a risk-coverage curve, reporting that our methods improve failure detection as well as trustworthiness of the model. The code are available at \url{https://github.com/AoShuang92/miscalibration_TS}.</description><subject>Artificial neural networks</subject><subject>Calibration</subject><subject>Confidence</subject><subject>Failure detection</subject><subject>Model accuracy</subject><subject>Safety critical</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>BENPR</sourceid><recordid>eNqNjE8LgjAYh0cQJOV3eKGzYJumdZWiDv2B7CzLvYuZbLVp0rfPIOja6Xd4nuc3IB5lbBakEaUj4jtXhWFI5wmNY-YRzDsDJyXQgZGwU67ktbpY3iijl7AVqBslX0pf4fBEC1wLOGuBNsiMln2mS4SjRaHKTwHSWNhj0xl7g-z3NCFDyWuH_nfHZLpe5dkmuFvzaNE1RWVaq3tU0DRK4oQlC8r-s95FyUa0</recordid><startdate>20230806</startdate><enddate>20230806</enddate><creator>Ao, Shuang</creator><creator>Rueger, Stefan</creator><creator>Siddharthan, Advaith</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>PTHSS</scope></search><sort><creationdate>20230806</creationdate><title>Two Sides of Miscalibration: Identifying Over and Under-Confidence Prediction for Network Calibration</title><author>Ao, Shuang ; Rueger, Stefan ; Siddharthan, Advaith</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_28475737923</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Artificial neural networks</topic><topic>Calibration</topic><topic>Confidence</topic><topic>Failure detection</topic><topic>Model accuracy</topic><topic>Safety critical</topic><toplevel>online_resources</toplevel><creatorcontrib>Ao, Shuang</creatorcontrib><creatorcontrib>Rueger, Stefan</creatorcontrib><creatorcontrib>Siddharthan, Advaith</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science &amp; Engineering Collection</collection><collection>ProQuest Central (Alumni Edition)</collection><collection>ProQuest Central UK/Ireland</collection><collection>ProQuest Central Essentials</collection><collection>ProQuest Central</collection><collection>Technology Collection (ProQuest)</collection><collection>ProQuest One Community College</collection><collection>ProQuest Central Korea</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>Engineering Collection</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Ao, Shuang</au><au>Rueger, Stefan</au><au>Siddharthan, Advaith</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Two Sides of Miscalibration: Identifying Over and Under-Confidence Prediction for Network Calibration</atitle><jtitle>arXiv.org</jtitle><date>2023-08-06</date><risdate>2023</risdate><eissn>2331-8422</eissn><abstract>Proper confidence calibration of deep neural networks is essential for reliable predictions in safety-critical tasks. Miscalibration can lead to model over-confidence and/or under-confidence; i.e., the model's confidence in its prediction can be greater or less than the model's accuracy. Recent studies have highlighted the over-confidence issue by introducing calibration techniques and demonstrated success on various tasks. However, miscalibration through under-confidence has not yet to receive much attention. In this paper, we address the necessity of paying attention to the under-confidence issue. We first introduce a novel metric, a miscalibration score, to identify the overall and class-wise calibration status, including being over or under-confident. Our proposed metric reveals the pitfalls of existing calibration techniques, where they often overly calibrate the model and worsen under-confident predictions. Then we utilize the class-wise miscalibration score as a proxy to design a calibration technique that can tackle both over and under-confidence. We report extensive experiments that show our proposed methods substantially outperforming existing calibration techniques. We also validate our proposed calibration technique on an automatic failure detection task with a risk-coverage curve, reporting that our methods improve failure detection as well as trustworthiness of the model. The code are available at \url{https://github.com/AoShuang92/miscalibration_TS}.</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, 2023-08
issn 2331-8422
language eng
recordid cdi_proquest_journals_2847573792
source Free E- Journals
subjects Artificial neural networks
Calibration
Confidence
Failure detection
Model accuracy
Safety critical
title Two Sides of Miscalibration: Identifying Over and Under-Confidence Prediction for Network Calibration
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-31T20%3A42%3A45IST&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=Two%20Sides%20of%20Miscalibration:%20Identifying%20Over%20and%20Under-Confidence%20Prediction%20for%20Network%20Calibration&rft.jtitle=arXiv.org&rft.au=Ao,%20Shuang&rft.date=2023-08-06&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E2847573792%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2847573792&rft_id=info:pmid/&rfr_iscdi=true