Style Feature Extraction Using Contrastive Conditioned Variational Autoencoders With Mutual Information Constraints

Extracting fine-grained features such as styles from unlabeled data is crucial for data analysis. Unsupervised methods such as variational autoencoders (VAEs) can extract styles that are usually mixed with other features. Conditional VAEs (CVAEs) can isolate styles using class labels; however, there...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE transactions on knowledge and data engineering Ročník 37; číslo 5; s. 3001 - 3014
Hlavní autoři: Yasutomi, Suguru, Tanaka, Toshihisa
Médium: Journal Article
Jazyk:angličtina
Vydáno: IEEE 01.05.2025
Témata:
ISSN:1041-4347, 1558-2191
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Abstract Extracting fine-grained features such as styles from unlabeled data is crucial for data analysis. Unsupervised methods such as variational autoencoders (VAEs) can extract styles that are usually mixed with other features. Conditional VAEs (CVAEs) can isolate styles using class labels; however, there are no established methods to extract only styles using unlabeled data. In this paper, we propose a CVAE-based method that extracts style features using only unlabeled data. The proposed model consists of a contrastive learning (CL) part that extracts style-independent features and a CVAE part that extracts style features. The CL model learns representations independent of data augmentation, which can be viewed as a perturbation in styles, in a self-supervised manner. Considering the style-independent features from the pretrained CL model as a condition, the CVAE learns to extract only styles. Additionally, we introduce a constraint based on mutual information between the CL and VAE features to prevent the CVAE from ignoring the condition. Experiments conducted using two simple datasets, MNIST and an original dataset based on Google Fonts, demonstrate that the proposed method can efficiently extract style features. Further experiments using real-world natural image datasets were also conducted to illustrate the method's extendability.
AbstractList Extracting fine-grained features such as styles from unlabeled data is crucial for data analysis. Unsupervised methods such as variational autoencoders (VAEs) can extract styles that are usually mixed with other features. Conditional VAEs (CVAEs) can isolate styles using class labels; however, there are no established methods to extract only styles using unlabeled data. In this paper, we propose a CVAE-based method that extracts style features using only unlabeled data. The proposed model consists of a contrastive learning (CL) part that extracts style-independent features and a CVAE part that extracts style features. The CL model learns representations independent of data augmentation, which can be viewed as a perturbation in styles, in a self-supervised manner. Considering the style-independent features from the pretrained CL model as a condition, the CVAE learns to extract only styles. Additionally, we introduce a constraint based on mutual information between the CL and VAE features to prevent the CVAE from ignoring the condition. Experiments conducted using two simple datasets, MNIST and an original dataset based on Google Fonts, demonstrate that the proposed method can efficiently extract style features. Further experiments using real-world natural image datasets were also conducted to illustrate the method's extendability.
Author Tanaka, Toshihisa
Yasutomi, Suguru
Author_xml – sequence: 1
  givenname: Suguru
  orcidid: 0009-0006-0626-982X
  surname: Yasutomi
  fullname: Yasutomi, Suguru
  email: yasutomi@sip.tuat.ac.jp
  organization: Department of Electronic and Information Engineering, Tokyo University of Agriculture and Technology, Koganei-Shi, Japan
– sequence: 2
  givenname: Toshihisa
  orcidid: 0000-0002-5056-9508
  surname: Tanaka
  fullname: Tanaka, Toshihisa
  organization: Department of Electronic and Information Engineering, Tokyo University of Agriculture and Technology, Koganei-Shi, Japan
BookMark eNpNkMtOAjEUhhuDiYA-gImLvsBgr8PMkiAgEeNC1OWk7ZzRGmhN2zHy9s4IC1fnz_kvi2-EBs47QOiakgmlpLzdPtwtJowwOeFScF7wMzSkUhYZoyUddJoImgkuphdoFOMnIaSYFnSI4nM67AAvQaU2AF78pKBMst7hl2jdO557131ist_Q69r2HtT4VQWreq12eNYmD874GkLEbzZ94Mc2tZ2xdo0P-79YX47dknUpXqLzRu0iXJ3uGG2Xi-38Pts8rdbz2SYzLOcpKxpJNFDeXamNqXPFaq1FrrQ2OQgucy4V00Y0jIqS5LVqtK451yyXipd8jOhx1gQfY4Cm-gp2r8KhoqTqoVU9tKqHVp2gdZ2bY8cCwL98UdJiKvgvTZJufw
CODEN ITKEEH
Cites_doi 10.5555/3524938.3525087
10.1145/1390156.1390294
10.1007/s11263-015-0816-y
10.1109/TVCG.2019.2921336
10.1109/CVPR42600.2020.00975
10.1109/ICCV.2015.425
10.1109/CVPR46437.2021.01549
10.1109/5.726791
10.1145/3581783.3612047
10.1109/CVPR.2016.90
10.1109/CVPR.2016.265
10.1109/ACCESS.2020.3031549
10.1371/journal.pone.0292126
10.1109/ICASSP39728.2021.9413528
10.5555/2969033.2969125
10.1109/TPAMI.2020.3031898
10.1109/ICASSP.2019.8683853
10.1109/IJCNN.2017.7966217
10.24963/ijcai.2021/324
10.1111/j.2517-6161.1948.tb00008.x
10.1109/CVPR.2018.00392
10.1016/j.neucom.2018.05.083
ContentType Journal Article
DBID 97E
ESBDL
RIA
RIE
AAYXX
CITATION
DOI 10.1109/TKDE.2025.3543383
DatabaseName IEEE Xplore (IEEE)
IEEE Xplore Open Access Journals
IEEE All-Society Periodicals Package (ASPP) 1998–Present
IEEE Electronic Library (IEL)
CrossRef
DatabaseTitle CrossRef
DatabaseTitleList
Database_xml – sequence: 1
  dbid: RIE
  name: IEEE Electronic Library (IEL)
  url: https://ieeexplore.ieee.org/
  sourceTypes: Publisher
DeliveryMethod fulltext_linktorsrc
Discipline Engineering
Computer Science
EISSN 1558-2191
EndPage 3014
ExternalDocumentID 10_1109_TKDE_2025_3543383
10891874
Genre orig-research
GrantInformation_xml – fundername: Japan Science and Technology Agency; CREST of the Japan Science and Technology Agency
  grantid: JPMJCR1784
  funderid: 10.13039/501100002241
GroupedDBID -~X
.DC
0R~
1OL
29I
4.4
5GY
5VS
6IK
97E
9M8
AAJGR
AARMG
AASAJ
AAWTH
ABAZT
ABFSI
ABQJQ
ABVLG
ACGFO
ACIWK
AENEX
AETIX
AGQYO
AGSQL
AHBIQ
AI.
AIBXA
AKJIK
AKQYR
ALLEH
ALMA_UNASSIGNED_HOLDINGS
ASUFR
ATWAV
BEFXN
BFFAM
BGNUA
BKEBE
BPEOZ
CS3
DU5
E.L
EBS
EJD
ESBDL
F5P
HZ~
H~9
ICLAB
IEDLZ
IFIPE
IFJZH
IPLJI
JAVBF
LAI
M43
MS~
O9-
OCL
P2P
PQQKQ
RIA
RIE
RNI
RNS
RXW
RZB
TAE
TAF
TN5
UHB
VH1
AAYXX
CITATION
ID FETCH-LOGICAL-c263t-8f50be138f55bccd6a2dbb46abbc6e435635a2bc4f214906dafbbd33b265a393
IEDL.DBID RIE
ISICitedReferencesCount 0
ISICitedReferencesURI http://www.webofscience.com/api/gateway?GWVersion=2&SrcApp=Summon&SrcAuth=ProQuest&DestLinkType=CitingArticles&DestApp=WOS_CPL&KeyUT=001459544100033&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
ISSN 1041-4347
IngestDate Sat Nov 29 08:05:21 EST 2025
Wed Aug 27 02:04:16 EDT 2025
IsDoiOpenAccess true
IsOpenAccess true
IsPeerReviewed true
IsScholarly true
Issue 5
Language English
License https://creativecommons.org/licenses/by/4.0/legalcode
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-c263t-8f50be138f55bccd6a2dbb46abbc6e435635a2bc4f214906dafbbd33b265a393
ORCID 0000-0002-5056-9508
0009-0006-0626-982X
OpenAccessLink https://ieeexplore.ieee.org/document/10891874
PageCount 14
ParticipantIDs crossref_primary_10_1109_TKDE_2025_3543383
ieee_primary_10891874
PublicationCentury 2000
PublicationDate 2025-05-01
PublicationDateYYYYMMDD 2025-05-01
PublicationDate_xml – month: 05
  year: 2025
  text: 2025-05-01
  day: 01
PublicationDecade 2020
PublicationTitle IEEE transactions on knowledge and data engineering
PublicationTitleAbbrev TKDE
PublicationYear 2025
Publisher IEEE
Publisher_xml – name: IEEE
References ref15
ref37
Chen (ref38) 2020
ref30
Kingma (ref41)
Goodfellow (ref3) 2016
ref33
Hjelm (ref35)
Howard (ref19) 2019
Katoh (ref20)
ref39
ref16
Kingma (ref13)
ref18
Bishop (ref1) 2006; 4
Mescheder (ref44)
Bardes (ref28)
Makhzani (ref23)
Higgins (ref24)
Sohn (ref14)
ref46
Kingma (ref12)
ref45
ref25
ref47
ref42
ref22
ref21
ref29
Murphy (ref2) 2012
ref8
ref7
Mescheder (ref43)
ref9
Zimmermann (ref11)
ref4
ref6
ref5
Long (ref32)
Locatello (ref26)
von Kügelgen (ref10)
Dugas (ref36)
ref40
Caron (ref27)
Belghazi (ref34)
Ganin (ref31) 2016; 17
References_xml – ident: ref6
  doi: 10.5555/3524938.3525087
– volume-title: Proc. Int. Conf. Learn. Representations
  ident: ref28
  article-title: VICReg: Variance-invariance-covariance regularization for self-supervised learning
– ident: ref4
  doi: 10.1145/1390156.1390294
– volume-title: Proc. Int. Conf. Learn. Representations
  ident: ref35
  article-title: Learning deep representations by mutual information estimation and maximization
– year: 2019
  ident: ref19
  article-title: Imagenette: A smaller subset of 10 easily classified classes from Imagenet
– ident: ref9
  doi: 10.1007/s11263-015-0816-y
– ident: ref22
  doi: 10.1109/TVCG.2019.2921336
– ident: ref7
  doi: 10.1109/CVPR42600.2020.00975
– start-page: 55
  volume-title: Proc. 2nd Int. Conf. Learn. Representations
  ident: ref12
  article-title: Auto-encoding variational bayes
– start-page: 3483
  volume-title: Proc. Adv. Neural Inf. Process. Syst.
  ident: ref14
  article-title: Learning structured output representation using deep conditional generative models
– start-page: 4114
  volume-title: Proc. 36th Int. Conf. Mach. Learn.
  ident: ref26
  article-title: Challenging common assumptions in the unsupervised learning of disentangled representations
– start-page: 16451
  volume-title: Proc. Adv. Neural Inf. Process. Syst.
  ident: ref10
  article-title: Self-supervised learning with data augmentations provably isolates content from style
– volume: 4
  volume-title: Pattern Recognition and Machine Learning
  year: 2006
  ident: ref1
– ident: ref18
  doi: 10.1109/ICCV.2015.425
– start-page: 9912
  volume-title: Proc. Adv. Neural Inf. Process. Syst.
  ident: ref27
  article-title: Unsupervised learning of visual features by contrasting cluster assignments
– volume-title: Machine Learning: A Probabilistic Perspective
  year: 2012
  ident: ref2
– volume-title: Proc. ICLR 2021 Workshop Generalization Beyond Training Distrib. Brains Mach.
  ident: ref20
  article-title: Dataset of annotated images of sundry objects — benchmark for performance degradation caused by domain shifts
– ident: ref8
  doi: 10.1109/CVPR46437.2021.01549
– volume-title: Proc. 3rd Int. Conf. Learn. Representations
  ident: ref41
  article-title: Adam: A method for stochastic optimization
– ident: ref15
  doi: 10.1109/5.726791
– volume-title: Proc. 4th Int. Conf. Learn. Representations
  ident: ref23
  article-title: Adversarial autoencoders
– start-page: 3581
  volume-title: Proc. Adv. Neural Inf. Process. Syst.
  ident: ref13
  article-title: Semi-supervised learning with deep generative models
– year: 2020
  ident: ref38
  article-title: Improved baselines with momentum contrastive learning
– start-page: 12979
  volume-title: Proc. 38th Int. Conf. Mach. Learn.
  ident: ref11
  article-title: Contrastive learning inverts the data generating process
– ident: ref29
  doi: 10.1145/3581783.3612047
– ident: ref40
  doi: 10.1109/CVPR.2016.90
– ident: ref21
  doi: 10.1109/CVPR.2016.265
– volume-title: Proc. 5th Int. Conf. Learn. Representations
  ident: ref24
  article-title: beta-VAE: Learning basic visual concepts with a constrained variational framework
– start-page: 136
  volume-title: Proc. Adv. Neural Inf. Process. Syst.
  ident: ref32
  article-title: Unsupervised domain adaptation with residual transfer networks
– start-page: 1823
  volume-title: Proc. Adv. Neural Inf. Process. Syst.
  ident: ref43
  article-title: The numerics of GANs
– ident: ref5
  doi: 10.1109/ACCESS.2020.3031549
– ident: ref45
  doi: 10.1371/journal.pone.0292126
– ident: ref46
  doi: 10.1109/ICASSP39728.2021.9413528
– ident: ref37
  doi: 10.5555/2969033.2969125
– start-page: 3481
  volume-title: Proc. 35th Int. Conf. Mach. Learn.
  ident: ref44
  article-title: Which training methods for GANs do actually converge?
– ident: ref25
  doi: 10.1109/TPAMI.2020.3031898
– ident: ref16
  doi: 10.1109/ICASSP.2019.8683853
– ident: ref39
  doi: 10.1109/IJCNN.2017.7966217
– ident: ref47
  doi: 10.24963/ijcai.2021/324
– start-page: 531
  volume-title: Proc. 35th Int. Conf. Mach. Learn.
  ident: ref34
  article-title: Mutual information neural estimation
– ident: ref42
  doi: 10.1111/j.2517-6161.1948.tb00008.x
– ident: ref33
  doi: 10.1109/CVPR.2018.00392
– volume-title: Deep Learning
  year: 2016
  ident: ref3
– ident: ref30
  doi: 10.1016/j.neucom.2018.05.083
– volume: 17
  start-page: 1
  issue: 59
  year: 2016
  ident: ref31
  article-title: Domain-adversarial training of neural networks
  publication-title: J. Mach. Learn. Res.
– start-page: 472
  volume-title: Proc. Adv. Neural Inf. Process. Syst.
  ident: ref36
  article-title: Incorporating second-order functional knowledge for better option pricing
SSID ssj0008781
Score 2.46151
Snippet Extracting fine-grained features such as styles from unlabeled data is crucial for data analysis. Unsupervised methods such as variational autoencoders (VAEs)...
SourceID crossref
ieee
SourceType Index Database
Publisher
StartPage 3001
SubjectTerms Autoencoders
contrastive learning
Data augmentation
Data mining
Data models
Feature extraction
Mutual information
Perturbation methods
Style extraction
Training
Unsupervised learning
variational autoencoders
Vectors
Title Style Feature Extraction Using Contrastive Conditioned Variational Autoencoders With Mutual Information Constraints
URI https://ieeexplore.ieee.org/document/10891874
Volume 37
WOSCitedRecordID wos001459544100033&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVIEE
  databaseName: IEEE Electronic Library (IEL)
  customDbUrl:
  eissn: 1558-2191
  dateEnd: 99991231
  omitProxy: false
  ssIdentifier: ssj0008781
  issn: 1041-4347
  databaseCode: RIE
  dateStart: 19890101
  isFulltext: true
  titleUrlDefault: https://ieeexplore.ieee.org/
  providerName: IEEE
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV3NS8MwFA86POjB6Zw4v8jBk9DZJm3SHoduCOoQHLpbyVdxIJt0qeh_b17ayTx48NRQmja81-S9l7zf-yF0QblOlBA6SAXXQUzjOJBMqcCEkonQmWgjvKbv-XicTqfZYwNW91gYY4xPPjN9aPqzfL1QFWyVuRmeZsAht4k2OWc1WOtn2U25ZyR14UUE3-TNEWYUZleTu5uhCwVJ0qdJDDHZLyO0xqrijcqo_c_h7KHdxnvEg1rd-2jDzDuovWJmwM1E7aCdtTKDB2j5ZL_eDAZvryoNHn7asoYzYJ8wgKFCVSmWsPBBW9flizR-dnF0s1eIB5VdQM1LyHvGLzP7ih8qgJ7gBs_kXwfsn55zwi67aDIaTq5vg4ZsIVCEURukRRJKE1F3TaRSmgmipYyZkFIx45wq55kIIlVcEBdUhUyLQkpNqSQsETSjh6g1d2M7Qtj1jrRKIm4KERtWZJEwPGWUSCLSNGM9dLkSfv5el9TIfSgSZjloKgdN5Y2meqgLgl97sJb58R_3T9A2dK9TEk9Ry5aVOUNb6sPOluW5_2O-Ab-cwx8
linkProvider IEEE
linkToHtml http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV1LS8QwEB58gXrwLb7NwZPQtU3StD2Krij7QHDRvZW8ioKodFvRf28mrbIePHhqKG0aZprMfMnMfAAnLDGxltIEqUxMwBnngRJaBzZUQobORFvpNd1PhsN0PM5u22R1nwtjrfXBZ7aDTX-Wb151jVtlboanGXLIzcJ8zDkNm3Stn4U3TTwnqQMYEX41aQ8xozA7G_Uuuw4M0rjDYo6o7JcZmuJV8WblavWfA1qDldZ_JOeNwtdhxr5swOo3NwNpp-oGLE8VGtyEyV31-WwJ-nt1aUn3oyqbhAbiQwYI1qgq5QSXPmybpoCRIfcOSbe7heS8rl6x6iVGPpOHp-qRDGpMPiFtRpPvDvk_PetENdmC0VV3dHEdtHQLgaaCVUFaxKGyEXPXWGlthKRGKS6kUlpY51Y530RSpXlBHawKhZGFUoYxRUUsWca2Ye7FjW0HiHs7MjqOEltIbkWRRdImqWBUUZmmmdiF02_h529NUY3cg5Ewy1FTOWoqbzW1C1so-KkHG5nv_XH_GBavR4N-3r8Z9vZhCbtqAhQPYK4qa3sIC_q9epqUR_7v-QINd8Zm
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%3Ajournal&rft.genre=article&rft.atitle=Style+Feature+Extraction+Using+Contrastive+Conditioned+Variational+Autoencoders+With+Mutual+Information+Constraints&rft.jtitle=IEEE+transactions+on+knowledge+and+data+engineering&rft.au=Yasutomi%2C+Suguru&rft.au=Tanaka%2C+Toshihisa&rft.date=2025-05-01&rft.issn=1041-4347&rft.eissn=1558-2191&rft.volume=37&rft.issue=5&rft.spage=3001&rft.epage=3014&rft_id=info:doi/10.1109%2FTKDE.2025.3543383&rft.externalDBID=n%2Fa&rft.externalDocID=10_1109_TKDE_2025_3543383
thumbnail_l http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=1041-4347&client=summon
thumbnail_m http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=1041-4347&client=summon
thumbnail_s http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=1041-4347&client=summon