Knowledge Accumulation in Continually Learned Representations and the Issue of Feature Forgetting

Continual learning research has shown that neural networks suffer from catastrophic forgetting "at the output level", but it is debated whether this is also the case at the level of learned representations. Multiple recent studies ascribe representations a certain level of innate robustnes...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:arXiv.org 2024-06
Hauptverfasser: Hess, Timm, Verwimp, Eli, Gido M van de Ven, Tuytelaars, Tinne
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 Hess, Timm
Verwimp, Eli
Gido M van de Ven
Tuytelaars, Tinne
description Continual learning research has shown that neural networks suffer from catastrophic forgetting "at the output level", but it is debated whether this is also the case at the level of learned representations. Multiple recent studies ascribe representations a certain level of innate robustness against forgetting -- that they only forget minimally in comparison with forgetting at the output level. We revisit and expand upon the experiments that revealed this difference in forgetting and illustrate the coexistence of two phenomena that affect the quality of continually learned representations: knowledge accumulation and feature forgetting. Taking both aspects into account, we show that, even though forgetting in the representation (i.e. feature forgetting) can be small in absolute terms, when measuring relative to how much was learned during a task, forgetting in the representation tends to be just as catastrophic as forgetting at the output level. Next we show that this feature forgetting is problematic as it substantially slows down the incremental learning of good general representations (i.e. knowledge accumulation). Finally, we study how feature forgetting and knowledge accumulation are affected by different types of continual learning methods.
format Article
fullrecord <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_2795077156</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2795077156</sourcerecordid><originalsourceid>FETCH-proquest_journals_27950771563</originalsourceid><addsrcrecordid>eNqNi8sKgkAUQIcgSMp_uNBa0LHRWoYkRa2ivQx6NWW6Y_Mg-vsk-oBWZ3HOmbGAp2kSbTecL1ho7RDHMc9yLkQaMHkm_VLYdAj7uvYPr6TrNUFPUGhyPXmp1BsuKA1hA1ccDVok960sSGrA3RFO1noE3UKJ0nmDUGrToZv-bsXmrVQWwx-XbF0ebsUxGo1-erSuGrQ3NKmK5zsR53kisvS_6gOICkXO</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2795077156</pqid></control><display><type>article</type><title>Knowledge Accumulation in Continually Learned Representations and the Issue of Feature Forgetting</title><source>Free E- Journals</source><creator>Hess, Timm ; Verwimp, Eli ; Gido M van de Ven ; Tuytelaars, Tinne</creator><creatorcontrib>Hess, Timm ; Verwimp, Eli ; Gido M van de Ven ; Tuytelaars, Tinne</creatorcontrib><description>Continual learning research has shown that neural networks suffer from catastrophic forgetting "at the output level", but it is debated whether this is also the case at the level of learned representations. Multiple recent studies ascribe representations a certain level of innate robustness against forgetting -- that they only forget minimally in comparison with forgetting at the output level. We revisit and expand upon the experiments that revealed this difference in forgetting and illustrate the coexistence of two phenomena that affect the quality of continually learned representations: knowledge accumulation and feature forgetting. Taking both aspects into account, we show that, even though forgetting in the representation (i.e. feature forgetting) can be small in absolute terms, when measuring relative to how much was learned during a task, forgetting in the representation tends to be just as catastrophic as forgetting at the output level. Next we show that this feature forgetting is problematic as it substantially slows down the incremental learning of good general representations (i.e. knowledge accumulation). Finally, we study how feature forgetting and knowledge accumulation are affected by different types of continual learning methods.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Accumulation ; Image classification ; Learning ; Neural networks ; Representations</subject><ispartof>arXiv.org, 2024-06</ispartof><rights>2024. This work is published under http://creativecommons.org/licenses/by/4.0/ (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>780,784</link.rule.ids></links><search><creatorcontrib>Hess, Timm</creatorcontrib><creatorcontrib>Verwimp, Eli</creatorcontrib><creatorcontrib>Gido M van de Ven</creatorcontrib><creatorcontrib>Tuytelaars, Tinne</creatorcontrib><title>Knowledge Accumulation in Continually Learned Representations and the Issue of Feature Forgetting</title><title>arXiv.org</title><description>Continual learning research has shown that neural networks suffer from catastrophic forgetting "at the output level", but it is debated whether this is also the case at the level of learned representations. Multiple recent studies ascribe representations a certain level of innate robustness against forgetting -- that they only forget minimally in comparison with forgetting at the output level. We revisit and expand upon the experiments that revealed this difference in forgetting and illustrate the coexistence of two phenomena that affect the quality of continually learned representations: knowledge accumulation and feature forgetting. Taking both aspects into account, we show that, even though forgetting in the representation (i.e. feature forgetting) can be small in absolute terms, when measuring relative to how much was learned during a task, forgetting in the representation tends to be just as catastrophic as forgetting at the output level. Next we show that this feature forgetting is problematic as it substantially slows down the incremental learning of good general representations (i.e. knowledge accumulation). Finally, we study how feature forgetting and knowledge accumulation are affected by different types of continual learning methods.</description><subject>Accumulation</subject><subject>Image classification</subject><subject>Learning</subject><subject>Neural networks</subject><subject>Representations</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>eNqNi8sKgkAUQIcgSMp_uNBa0LHRWoYkRa2ivQx6NWW6Y_Mg-vsk-oBWZ3HOmbGAp2kSbTecL1ho7RDHMc9yLkQaMHkm_VLYdAj7uvYPr6TrNUFPUGhyPXmp1BsuKA1hA1ccDVok960sSGrA3RFO1noE3UKJ0nmDUGrToZv-bsXmrVQWwx-XbF0ebsUxGo1-erSuGrQ3NKmK5zsR53kisvS_6gOICkXO</recordid><startdate>20240624</startdate><enddate>20240624</enddate><creator>Hess, Timm</creator><creator>Verwimp, Eli</creator><creator>Gido M van de Ven</creator><creator>Tuytelaars, Tinne</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>20240624</creationdate><title>Knowledge Accumulation in Continually Learned Representations and the Issue of Feature Forgetting</title><author>Hess, Timm ; Verwimp, Eli ; Gido M van de Ven ; Tuytelaars, Tinne</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_27950771563</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Accumulation</topic><topic>Image classification</topic><topic>Learning</topic><topic>Neural networks</topic><topic>Representations</topic><toplevel>online_resources</toplevel><creatorcontrib>Hess, Timm</creatorcontrib><creatorcontrib>Verwimp, Eli</creatorcontrib><creatorcontrib>Gido M van de Ven</creatorcontrib><creatorcontrib>Tuytelaars, Tinne</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>Hess, Timm</au><au>Verwimp, Eli</au><au>Gido M van de Ven</au><au>Tuytelaars, Tinne</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Knowledge Accumulation in Continually Learned Representations and the Issue of Feature Forgetting</atitle><jtitle>arXiv.org</jtitle><date>2024-06-24</date><risdate>2024</risdate><eissn>2331-8422</eissn><abstract>Continual learning research has shown that neural networks suffer from catastrophic forgetting "at the output level", but it is debated whether this is also the case at the level of learned representations. Multiple recent studies ascribe representations a certain level of innate robustness against forgetting -- that they only forget minimally in comparison with forgetting at the output level. We revisit and expand upon the experiments that revealed this difference in forgetting and illustrate the coexistence of two phenomena that affect the quality of continually learned representations: knowledge accumulation and feature forgetting. Taking both aspects into account, we show that, even though forgetting in the representation (i.e. feature forgetting) can be small in absolute terms, when measuring relative to how much was learned during a task, forgetting in the representation tends to be just as catastrophic as forgetting at the output level. Next we show that this feature forgetting is problematic as it substantially slows down the incremental learning of good general representations (i.e. knowledge accumulation). Finally, we study how feature forgetting and knowledge accumulation are affected by different types of continual learning methods.</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_2795077156
source Free E- Journals
subjects Accumulation
Image classification
Learning
Neural networks
Representations
title Knowledge Accumulation in Continually Learned Representations and the Issue of Feature Forgetting
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-04T12%3A29%3A11IST&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=Knowledge%20Accumulation%20in%20Continually%20Learned%20Representations%20and%20the%20Issue%20of%20Feature%20Forgetting&rft.jtitle=arXiv.org&rft.au=Hess,%20Timm&rft.date=2024-06-24&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E2795077156%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2795077156&rft_id=info:pmid/&rfr_iscdi=true