Online Class Incremental Learning with One-Vs-All Classifiers for Resource Constrained Devices

Online Class Incremental Learning (OCIL) aims to learn new classes from a data stream where samples arrive in batches, one after the other. Avoiding catastrophic forgetting, the phenomenon of forgetting old classes when learning new ones is the main challenge in OCIL. Replay-based methods counteract...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2023 International Symposium on Image and Signal Processing and Analysis (ISPA) s. 1 - 6
Hlavní autoři: Baptiste, Wagner, Denis, Pellerin, Serge, Olympieff, Sylvain, Huet
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 18.09.2023
Témata:
ISSN:1849-2266
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Abstract Online Class Incremental Learning (OCIL) aims to learn new classes from a data stream where samples arrive in batches, one after the other. Avoiding catastrophic forgetting, the phenomenon of forgetting old classes when learning new ones is the main challenge in OCIL. Replay-based methods counteract catastrophic forgetting by storing around 10% of the data stream in a memory buffer. Upon learning new classes, the model is updated by replaying old class images sampled from memory. OCIL holds significant promise for smart devices, such as home robots or smartphones, as incrementally learning new object instances enables personalized interactions with the environment. Although, these devices present limited computing and storage capabilities to allow on-device training in real-time. In this paper, we propose a novel replay-based method called ILOVA (Incremental Learning of One-Vs-All classifiers) and show that it achieves the best balance between accuracy, forgetting, computing time, and memory footprint on three benchmark datasets. Additionally, we conduct a comparative analysis of existing replay-based methods for OCIL with respect to embedded constraints. Specifically in the studied scenarios, models can store only one to ten samples per class. In the most challenging configuration, where only one sample per class is stored, our method outperforms the second-best method by up to 16 points in accuracy within 2.5 times less computation time.
AbstractList Online Class Incremental Learning (OCIL) aims to learn new classes from a data stream where samples arrive in batches, one after the other. Avoiding catastrophic forgetting, the phenomenon of forgetting old classes when learning new ones is the main challenge in OCIL. Replay-based methods counteract catastrophic forgetting by storing around 10% of the data stream in a memory buffer. Upon learning new classes, the model is updated by replaying old class images sampled from memory. OCIL holds significant promise for smart devices, such as home robots or smartphones, as incrementally learning new object instances enables personalized interactions with the environment. Although, these devices present limited computing and storage capabilities to allow on-device training in real-time. In this paper, we propose a novel replay-based method called ILOVA (Incremental Learning of One-Vs-All classifiers) and show that it achieves the best balance between accuracy, forgetting, computing time, and memory footprint on three benchmark datasets. Additionally, we conduct a comparative analysis of existing replay-based methods for OCIL with respect to embedded constraints. Specifically in the studied scenarios, models can store only one to ten samples per class. In the most challenging configuration, where only one sample per class is stored, our method outperforms the second-best method by up to 16 points in accuracy within 2.5 times less computation time.
Author Serge, Olympieff
Baptiste, Wagner
Denis, Pellerin
Sylvain, Huet
Author_xml – sequence: 1
  givenname: Wagner
  surname: Baptiste
  fullname: Baptiste, Wagner
  organization: Univ. Grenoble Alpes, CNRS, Grenoble INP,GIPSA-lab,Grenoble,France,38000
– sequence: 2
  givenname: Pellerin
  surname: Denis
  fullname: Denis, Pellerin
  organization: Univ. Grenoble Alpes, CNRS, Grenoble INP,GIPSA-lab,Grenoble,France,38000
– sequence: 3
  givenname: Olympieff
  surname: Serge
  fullname: Serge, Olympieff
  organization: Univ. Grenoble Alpes, CNRS, Grenoble INP,GIPSA-lab,Grenoble,France,38000
– sequence: 4
  givenname: Huet
  surname: Sylvain
  fullname: Sylvain, Huet
  organization: Univ. Grenoble Alpes, CNRS, Grenoble INP,GIPSA-lab,Grenoble,France,38000
BookMark eNo1kM1KAzEURqMoWGvfQDAvMDU3dzKTLEv9KxQq_i0tmcyNBqYZSUbFt3eguvrO4nAW3yk7in0kxi5AzAGEuVw93i-URgVzKSTOQcjaaFkdsJkZAZVAUFjhIZuALk0hZVWdsFnOoRGlVqIcpQl73cQuROLLzubMV9El2lEcbMfXZFMM8Y1_h-GdbyIVL7lYdN1eDT5Qytz3iT9Q7j-TGxt9zEOyY67lV_QVHOUzduxtl2n2t1P2fHP9tLwr1pvb1XKxLgKAGQpPnpxujPU1GiuUdhKpdtbVCsu2Mg1K5SyCboQH6QyqlipLTesBalCEU3a-7wYi2n6ksLPpZ_t_Cf4CkxVZMg
ContentType Conference Proceeding
DBID 6IE
6IL
CBEJK
RIE
RIL
DOI 10.1109/ISPA58351.2023.10279826
DatabaseName IEEE Electronic Library (IEL) Conference Proceedings
IEEE Xplore POP ALL
IEEE Xplore All Conference Proceedings
IEEE/IET Electronic Library
IEEE Proceedings Order Plans (POP All) 1998-Present
DatabaseTitleList
Database_xml – sequence: 1
  dbid: RIE
  name: IEEE/IET Electronic Library
  url: https://ieeexplore.ieee.org/
  sourceTypes: Publisher
DeliveryMethod fulltext_linktorsrc
EISBN 9798350315363
EISSN 1849-2266
EndPage 6
ExternalDocumentID 10279826
Genre orig-research
GroupedDBID 6IE
6IL
ABLEC
ALMA_UNASSIGNED_HOLDINGS
CBEJK
IEGSK
RIE
RIL
ID FETCH-LOGICAL-i119t-fefec8b9af739a058c23e7cac7534d69b325ca318b0f12c935de6aebdf11715e3
IEDL.DBID RIE
IngestDate Wed Jun 26 19:24:08 EDT 2024
IsPeerReviewed false
IsScholarly false
Language English
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-i119t-fefec8b9af739a058c23e7cac7534d69b325ca318b0f12c935de6aebdf11715e3
PageCount 6
ParticipantIDs ieee_primary_10279826
PublicationCentury 2000
PublicationDate 2023-Sept.-18
PublicationDateYYYYMMDD 2023-09-18
PublicationDate_xml – month: 09
  year: 2023
  text: 2023-Sept.-18
  day: 18
PublicationDecade 2020
PublicationTitle 2023 International Symposium on Image and Signal Processing and Analysis (ISPA)
PublicationTitleAbbrev ISPA
PublicationYear 2023
Publisher IEEE
Publisher_xml – name: IEEE
SSID ssib048504798
ssib042470063
Score 1.8524514
Snippet Online Class Incremental Learning (OCIL) aims to learn new classes from a data stream where samples arrive in batches, one after the other. Avoiding...
SourceID ieee
SourceType Publisher
StartPage 1
SubjectTerms Benchmark testing
Catas-trophic forgetting
Computational modeling
Incremental learning
Online learning
Real-time systems
Replay
Signal processing
Stability analysis
Streaming media
Training
Title Online Class Incremental Learning with One-Vs-All Classifiers for Resource Constrained Devices
URI https://ieeexplore.ieee.org/document/10279826
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV07T8MwELZoxcAEiCLe8sDqEttxbI8VUMFSKvFQJyo_LqhSlaI--P3YTlLEwMAWR5Fl-XK5z5e770PomhkbUDLPiC60JbnnObGWOSKt8swUVpbGJbEJORqpyUSPm2b11AsDAKn4DPrxMv3L9wu3iamy4OFM6oCHO6gjZVE3a7UvT85yGePtdqxEZE9XTU0XzfTN4_N4IALiiMdCxvvtbL90VVJYGe7_c0EHqPfToIfH29BziHagOkLvNW0oTjqXODh-nfozc9yQqH7gmHXFTxWQtxUZzOf1o7MyymHjgF5xm8zHUcczqUeAx3eQviY99Dq8f7l9II18AplRqtekhBKcstqUkmuTCeUYB-mMCyeU3AfTcCacCT5ts5Iyp7nwUBiwvqRUUgH8GHWrRQUnCBfBahFYMS98blxhrfCKmcg6HuYBf4p6cXOmnzVDxrTdl7M_7p-jvWiCWHdB1QXqrpcbuES77ms9Wy2vkl2_AR41pAE
linkProvider IEEE
linkToHtml http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV3JTsMwELWgIMEJEEXs-MDVJXbsJD5WLGpFKZUoqCcqLxNUqUpRF74f20mKOHDgFkeRZXkymefJzHsIXTOlHUqOIyITqQm3MSdaM0NSnVmmEp3mygSxibTfz0YjOaia1UMvDACE4jNo-cvwL9_OzMqnypyHs1Q6PLyJtgTnLCrbterXhzOe-oi7HmfC86dnVVUXjeRN92XQFg5z-IMhi1v1fL-UVUJgedj755L2UfOnRQ8P1sHnAG1AcYjeS-JQHJQusXP9MvmnpriiUf3APu-KnwsgbwvSnk7LRye5F8TGDr_iOp2PvZJn0I8Ai-8gfE-a6PXhfnjbIZWAAplQKpckhxxMpqXK01iqSGSGxZAaZdwZhVtnnJgJo5xX6yinzMhYWEgUaJtTmlIB8RFqFLMCjhFOnN08tGJWWK5MorWwGVOed9zNA_YENf3mjD9LjoxxvS-nf9y_Qjud4VNv3Ov2H8_QrjeHr8Kg2TlqLOcruEDb5ms5Wcwvg42_AdTtp0g
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=proceeding&rft.title=2023+International+Symposium+on+Image+and+Signal+Processing+and+Analysis+%28ISPA%29&rft.atitle=Online+Class+Incremental+Learning+with+One-Vs-All+Classifiers+for+Resource+Constrained+Devices&rft.au=Baptiste%2C+Wagner&rft.au=Denis%2C+Pellerin&rft.au=Serge%2C+Olympieff&rft.au=Sylvain%2C+Huet&rft.date=2023-09-18&rft.pub=IEEE&rft.eissn=1849-2266&rft.spage=1&rft.epage=6&rft_id=info:doi/10.1109%2FISPA58351.2023.10279826&rft.externalDocID=10279826