Understanding Self-Distillation and Partial Label Learning in Multi-Class Classification with Label Noise

Self-distillation (SD) is the process of training a student model using the outputs of a teacher model, with both models sharing the same architecture. Our study theoretically examines SD in multi-class classification with cross-entropy loss, exploring both multi-round SD and SD with refined teacher...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:arXiv.org 2024-02
Hauptverfasser: Jeong, Hyeonsu, Chung, Hye Won
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 Jeong, Hyeonsu
Chung, Hye Won
description Self-distillation (SD) is the process of training a student model using the outputs of a teacher model, with both models sharing the same architecture. Our study theoretically examines SD in multi-class classification with cross-entropy loss, exploring both multi-round SD and SD with refined teacher outputs, inspired by partial label learning (PLL). By deriving a closed-form solution for the student model's outputs, we discover that SD essentially functions as label averaging among instances with high feature correlations. Initially beneficial, this averaging helps the model focus on feature clusters correlated with a given instance for predicting the label. However, it leads to diminishing performance with increasing distillation rounds. Additionally, we demonstrate SD's effectiveness in label noise scenarios and identify the label corruption condition and minimum number of distillation rounds needed to achieve 100% classification accuracy. Our study also reveals that one-step distillation with refined teacher outputs surpasses the efficacy of multi-step SD using the teacher's direct output in high noise rate regimes.
format Article
fullrecord <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_2928440253</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2928440253</sourcerecordid><originalsourceid>FETCH-proquest_journals_29284402533</originalsourceid><addsrcrecordid>eNqNi8sKwjAQRYMgWLT_EHAdqEmrdV0VFyqCui6jTXVKSDST4u9bHx_g5tzFObfHIqnUROSplAMWEzVJksjpTGaZihiebKU9BbAV2is_aFOLBVJAYyCgs7wTfA8-IBi-gbPuqMHbd4yWb1sTUBQGiPiHWOPle3xiuP0eO4ekR6xfgyEd_3bIxqvlsViLu3ePVlMoG9d626lSzmWeponMlPqvegGTw0f0</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2928440253</pqid></control><display><type>article</type><title>Understanding Self-Distillation and Partial Label Learning in Multi-Class Classification with Label Noise</title><source>Free E- Journals</source><creator>Jeong, Hyeonsu ; Chung, Hye Won</creator><creatorcontrib>Jeong, Hyeonsu ; Chung, Hye Won</creatorcontrib><description>Self-distillation (SD) is the process of training a student model using the outputs of a teacher model, with both models sharing the same architecture. Our study theoretically examines SD in multi-class classification with cross-entropy loss, exploring both multi-round SD and SD with refined teacher outputs, inspired by partial label learning (PLL). By deriving a closed-form solution for the student model's outputs, we discover that SD essentially functions as label averaging among instances with high feature correlations. Initially beneficial, this averaging helps the model focus on feature clusters correlated with a given instance for predicting the label. However, it leads to diminishing performance with increasing distillation rounds. Additionally, we demonstrate SD's effectiveness in label noise scenarios and identify the label corruption condition and minimum number of distillation rounds needed to achieve 100% classification accuracy. Our study also reveals that one-step distillation with refined teacher outputs surpasses the efficacy of multi-step SD using the teacher's direct output in high noise rate regimes.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Classification ; Distillation ; Labels ; Learning ; Machine learning ; Teachers</subject><ispartof>arXiv.org, 2024-02</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>Jeong, Hyeonsu</creatorcontrib><creatorcontrib>Chung, Hye Won</creatorcontrib><title>Understanding Self-Distillation and Partial Label Learning in Multi-Class Classification with Label Noise</title><title>arXiv.org</title><description>Self-distillation (SD) is the process of training a student model using the outputs of a teacher model, with both models sharing the same architecture. Our study theoretically examines SD in multi-class classification with cross-entropy loss, exploring both multi-round SD and SD with refined teacher outputs, inspired by partial label learning (PLL). By deriving a closed-form solution for the student model's outputs, we discover that SD essentially functions as label averaging among instances with high feature correlations. Initially beneficial, this averaging helps the model focus on feature clusters correlated with a given instance for predicting the label. However, it leads to diminishing performance with increasing distillation rounds. Additionally, we demonstrate SD's effectiveness in label noise scenarios and identify the label corruption condition and minimum number of distillation rounds needed to achieve 100% classification accuracy. Our study also reveals that one-step distillation with refined teacher outputs surpasses the efficacy of multi-step SD using the teacher's direct output in high noise rate regimes.</description><subject>Classification</subject><subject>Distillation</subject><subject>Labels</subject><subject>Learning</subject><subject>Machine learning</subject><subject>Teachers</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>eNqNi8sKwjAQRYMgWLT_EHAdqEmrdV0VFyqCui6jTXVKSDST4u9bHx_g5tzFObfHIqnUROSplAMWEzVJksjpTGaZihiebKU9BbAV2is_aFOLBVJAYyCgs7wTfA8-IBi-gbPuqMHbd4yWb1sTUBQGiPiHWOPle3xiuP0eO4ekR6xfgyEd_3bIxqvlsViLu3ePVlMoG9d626lSzmWeponMlPqvegGTw0f0</recordid><startdate>20240216</startdate><enddate>20240216</enddate><creator>Jeong, Hyeonsu</creator><creator>Chung, Hye Won</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>20240216</creationdate><title>Understanding Self-Distillation and Partial Label Learning in Multi-Class Classification with Label Noise</title><author>Jeong, Hyeonsu ; Chung, Hye Won</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_29284402533</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Classification</topic><topic>Distillation</topic><topic>Labels</topic><topic>Learning</topic><topic>Machine learning</topic><topic>Teachers</topic><toplevel>online_resources</toplevel><creatorcontrib>Jeong, Hyeonsu</creatorcontrib><creatorcontrib>Chung, Hye Won</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</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>ProQuest Central China</collection><collection>Engineering Collection</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Jeong, Hyeonsu</au><au>Chung, Hye Won</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Understanding Self-Distillation and Partial Label Learning in Multi-Class Classification with Label Noise</atitle><jtitle>arXiv.org</jtitle><date>2024-02-16</date><risdate>2024</risdate><eissn>2331-8422</eissn><abstract>Self-distillation (SD) is the process of training a student model using the outputs of a teacher model, with both models sharing the same architecture. Our study theoretically examines SD in multi-class classification with cross-entropy loss, exploring both multi-round SD and SD with refined teacher outputs, inspired by partial label learning (PLL). By deriving a closed-form solution for the student model's outputs, we discover that SD essentially functions as label averaging among instances with high feature correlations. Initially beneficial, this averaging helps the model focus on feature clusters correlated with a given instance for predicting the label. However, it leads to diminishing performance with increasing distillation rounds. Additionally, we demonstrate SD's effectiveness in label noise scenarios and identify the label corruption condition and minimum number of distillation rounds needed to achieve 100% classification accuracy. Our study also reveals that one-step distillation with refined teacher outputs surpasses the efficacy of multi-step SD using the teacher's direct output in high noise rate regimes.</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, 2024-02
issn 2331-8422
language eng
recordid cdi_proquest_journals_2928440253
source Free E- Journals
subjects Classification
Distillation
Labels
Learning
Machine learning
Teachers
title Understanding Self-Distillation and Partial Label Learning in Multi-Class Classification with Label Noise
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-26T08%3A54%3A40IST&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=Understanding%20Self-Distillation%20and%20Partial%20Label%20Learning%20in%20Multi-Class%20Classification%20with%20Label%20Noise&rft.jtitle=arXiv.org&rft.au=Jeong,%20Hyeonsu&rft.date=2024-02-16&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E2928440253%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2928440253&rft_id=info:pmid/&rfr_iscdi=true