A cosine similarity-based token subsampling method for vision transformer in cloud computing

Deploying huge deep learning applications on resource-constrained edge devices is a challenging task. Cloud-based edge computing is a promising solution. Such as model partitioning, a portion of the deep learning model is deployed on the edge device; while, the remaining portion is executed by the c...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Veröffentlicht in:	Neural computing & applications Jg. 37; H. 4; S. 2627 - 2639
Hauptverfasser:	Li, Qi, Kaneko, Hayata, Meng, Lin
Format:	Journal Article
Sprache:	Englisch
Veröffentlicht:	London Springer London 01.02.2025 Springer Nature B.V
Schlagworte:	Algorithms Artificial Intelligence Centroids Cloud computing Computational Biology/Bioinformatics Computational Science and Engineering Computer Science Data Mining and Knowledge Discovery Deep learning Edge computing Image Processing and Computer Vision Original Article Partitioning Probability and Statistics in Computer Science Similarity Transmission efficiency Vision Visual tasks Model partitioning Cosine similarity Vision transformer Cloud computing Token clustering Edge computing
ISSN:	0941-0643, 1433-3058
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Abstract	Deploying huge deep learning applications on resource-constrained edge devices is a challenging task. Cloud-based edge computing is a promising solution. Such as model partitioning, a portion of the deep learning model is deployed on the edge device; while, the remaining portion is executed by the cloud. Leveraging the computation power of edge devices, transmission latency is reduced, and bandwidth efficiency is increased. Recently, visual transformer models, supported by large datasets, have dominated in multiple vision tasks. However, model partitioning optimization methods for visual transformers are lacking. Therefore, the paper proposes a cosine similarity-based token subsampling method for visual transformer model partitioning to improve transmission efficiency. Tokens in the same class are subsampled and only the centroid tokens are uploaded. In the cloud, all tokens are reconstructed based on interpolation indexes. Three algorithm implementations are proposed and measured on PC, Jetson NANO and edge CPU Cortex-A53. The experimental results demonstrate that the recommended algorithm implementation can be executed with low-latency of 71.24 ms, and 35.65% transmitted data is reduced with an accuracy drop of 0.46%.
AbstractList	Deploying huge deep learning applications on resource-constrained edge devices is a challenging task. Cloud-based edge computing is a promising solution. Such as model partitioning, a portion of the deep learning model is deployed on the edge device; while, the remaining portion is executed by the cloud. Leveraging the computation power of edge devices, transmission latency is reduced, and bandwidth efficiency is increased. Recently, visual transformer models, supported by large datasets, have dominated in multiple vision tasks. However, model partitioning optimization methods for visual transformers are lacking. Therefore, the paper proposes a cosine similarity-based token subsampling method for visual transformer model partitioning to improve transmission efficiency. Tokens in the same class are subsampled and only the centroid tokens are uploaded. In the cloud, all tokens are reconstructed based on interpolation indexes. Three algorithm implementations are proposed and measured on PC, Jetson NANO and edge CPU Cortex-A53. The experimental results demonstrate that the recommended algorithm implementation can be executed with low-latency of 71.24 ms, and 35.65% transmitted data is reduced with an accuracy drop of 0.46%. Deploying huge deep learning applications on resource-constrained edge devices is a challenging task. Cloud-based edge computing is a promising solution. Such as model partitioning, a portion of the deep learning model is deployed on the edge device; while, the remaining portion is executed by the cloud. Leveraging the computation power of edge devices, transmission latency is reduced, and bandwidth efficiency is increased. Recently, visual transformer models, supported by large datasets, have dominated in multiple vision tasks. However, model partitioning optimization methods for visual transformers are lacking. Therefore, the paper proposes a cosine similarity-based token subsampling method for visual transformer model partitioning to improve transmission efficiency. Tokens in the same class are subsampled and only the centroid tokens are uploaded. In the cloud, all tokens are reconstructed based on interpolation indexes. Three algorithm implementations are proposed and measured on PC, Jetson NANO and edge CPU Cortex-A53. The experimental results demonstrate that the recommended algorithm implementation can be executed with low-latency of 71.24 ms, and 35.65% transmitted data is reduced with an accuracy drop of 0.46%.
Author	Meng, Lin Li, Qi Kaneko, Hayata
Author_xml	– sequence: 1 givenname: Qi surname: Li fullname: Li, Qi organization: Graduate School of Science and Engineering, Ritsumeikan University – sequence: 2 givenname: Hayata surname: Kaneko fullname: Kaneko, Hayata organization: Graduate School of Science and Engineering, Ritsumeikan University – sequence: 3 givenname: Lin orcidid: 0000-0003-4351-6923 surname: Meng fullname: Meng, Lin email: menglin@fc.ritsumei.ac.jp organization: College of Science and Engineering, Ritsumeikan University
BookMark	eNp9kE1LAzEQhoNUsK3-AU8Bz6uTj_06luIXFLzoTQjZ7GxN7SY12bX03xut4M3TMMzzvgPPjEycd0jIJYNrBlDeRICcswy4zBiUrMr2J2TKpBCZgLyakCnUMp0LKc7ILMYNAMiiyqfkdUGNj9Yhjba3Wx3scMgaHbGlg39HR-PYRN3vttataY_Dm29p5wP9tNF6R4egXUx7j4FaR83Wj20q7HfjkALn5LTT24gXv3NOXu5un5cP2erp_nG5WGWGl2KfFRo7JjSatu4k5pI33DSSaeRQNbwWvMa66XSJXCYIjCzbogAhAYVEY2oxJ1fH3l3wHyPGQW38GFx6qQQrRAGclVWi-JEywccYsFO7YHsdDoqB-raojhZVsqh-LKp9ColjKCbYrTH8Vf-T-gIq8nlv
Cites_doi	10.1186/s40537-021-00444-8 10.1145/216585.216588 10.1007/s00521-024-09910-9 10.1109/83.334991 10.3390/heritage6050230 10.1016/j.neunet.2022.06.038 10.23919/DATE.2017.7927211 10.1109/ICDCS.2017.226 10.1109/ACCESS.2021.3084689 10.1145/3093337.3037698 10.1007/s11263-015-0816-y 10.1109/TIE.2020.2998756 10.1109/TIM.2023.3264039 10.1109/ACCESS.2020.2991734 10.1109/TNET.2020.3042320 10.1109/TNNLS.2023.3240195 10.1145/3373087.3375887 10.1109/HPCA47549.2020.00036 10.1007/s00521-024-10015-6 10.1007/978-3-030-69538-5_4 10.1109/JPROC.2019.2921977 10.1007/978-3-030-89657-7_3 10.1007/3-540-57301-1_5 10.1109/JSTSP.2023.3239189 10.1016/j.eng.2020.01.007 10.1016/j.iot.2023.101047 10.1007/s00521-024-09891-9 10.1145/2809695.2809711 10.1145/3516807.3516820 10.3390/s23146547 10.1038/s41586-020-2649-2 10.1007/s10115-022-01776-4 10.1109/TCAD.2018.2858384 10.1109/ICCV48922.2021.00986 10.1109/TPAMI.2022.3152247 10.1007/s11554-024-01496-8 10.1109/ICCV48922.2021.00060 10.1109/INFOCOM.2018.8485905 10.1007/s00521-024-09938-x 10.1007/s00521-024-09838-0
ContentType	Journal Article
Copyright	The Author(s) 2024 Copyright Springer Nature B.V. Feb 2025
Copyright_xml	– notice: The Author(s) 2024 – notice: Copyright Springer Nature B.V. Feb 2025
DBID	C6C AAYXX CITATION
DOI	10.1007/s00521-024-10718-w
DatabaseName	Springer Nature OA/Free Journals CrossRef
DatabaseTitle	CrossRef
DatabaseTitleList	CrossRef
DeliveryMethod	fulltext_linktorsrc
Discipline	Computer Science
EISSN	1433-3058
EndPage	2639
ExternalDocumentID	10_1007_s00521_024_10718_w
GrantInformation_xml	– fundername: KIOXIA Holdings Corporation funderid: http://dx.doi.org/10.13039/100019355 – fundername: Ritsumeikan University
GroupedDBID	-Y2 -~C .4S .86 .DC .VR 06D 0R~ 0VY 123 1N0 1SB 2.D 203 28- 29N 2J2 2JN 2JY 2KG 2LR 2P1 2VQ 2~H 30V 4.4 406 408 409 40D 40E 53G 5QI 5VS 67Z 6NX 8FE 8FG 8TC 8UJ 95- 95. 95~ 96X AAAVM AABHQ AACDK AAHNG AAIAL AAJBT AAJKR AANZL AAOBN AAPKM AARHV AARTL AASML AATNV AATVU AAUYE AAWCG AAYIU AAYQN AAYTO AAYZH ABAKF ABBBX ABBRH ABBXA ABDBE ABDBF ABDZT ABECU ABFSG ABFTD ABFTV ABHLI ABHQN ABJNI ABJOX ABKCH ABKTR ABLJU ABMNI ABMQK ABNWP ABQBU ABQSL ABRTQ ABSXP ABTEG ABTHY ABTKH ABTMW ABULA ABWNU ABXPI ACAOD ACBXY ACDTI ACGFS ACHSB ACHXU ACKNC ACMDZ ACMLO ACOKC ACOMO ACPIV ACSNA ACSTC ACUHS ACZOJ ADHHG ADHIR ADHKG ADIMF ADKFA ADKNI ADKPE ADMLS ADRFC ADTPH ADURQ ADYFF ADZKW AEBTG AEFIE AEFQL AEGAL AEGNC AEJHL AEJRE AEKMD AEMSY AENEX AEOHA AEPYU AESKC AETLH AEVLU AEXYK AEZWR AFBBN AFDZB AFEXP AFGCZ AFHIU AFKRA AFLOW AFOHR AFQWF AFWTZ AFZKB AGAYW AGDGC AGGDS AGJBK AGMZJ AGQEE AGQMX AGQPQ AGRTI AGWIL AGWZB AGYKE AHAVH AHBYD AHKAY AHPBZ AHSBF AHWEU AHYZX AIAKS AIGIU AIIXL AILAN AITGF AIXLP AJBLW AJRNO AJZVZ ALMA_UNASSIGNED_HOLDINGS ALWAN AMKLP AMXSW AMYLF AMYQR AOCGG ARAPS ARCSS ARMRJ ASPBG ATHPR AVWKF AXYYD AYFIA AYJHY AZFZN B-. B0M BA0 BBWZM BDATZ BENPR BGLVJ BGNMA BSONS C6C CAG CCPQU COF CS3 CSCUP DDRTE DL5 DNIVK DPUIP DU5 EAD EAP EBLON EBS ECS EDO EIOEI EJD EMI EMK EPL ESBYG EST ESX F5P FEDTE FERAY FFXSO FIGPU FINBP FNLPD FRRFC FSGXE FWDCC GGCAI GGRSB GJIRD GNWQR GQ7 GQ8 GXS H13 HCIFZ HF~ HG5 HG6 HMJXF HQYDN HRMNR HVGLF HZ~ I-F I09 IHE IJ- IKXTQ ITM IWAJR IXC IZIGR IZQ I~X I~Z J-C J0Z JBSCW JCJTX JZLTJ KDC KOV KOW LAS LLZTM M4Y MA- N2Q N9A NB0 NDZJH NPVJJ NQJWS NU0 O9- O93 O9G O9I O9J OAM P19 P2P P62 P9O PF0 PHGZM PHGZT PQGLB PT4 PT5 QOK QOS R4E R89 R9I RHV RIG RNI RNS ROL RPX RSV RZK S16 S1Z S26 S27 S28 S3B SAP SCJ SCLPG SCO SDH SDM SHX SISQX SJYHP SNE SNPRN SNX SOHCF SOJ SPISZ SRMVM SSLCW STPWE SZN T13 T16 TSG TSK TSV TUC TUS U2A UG4 UOJIU UTJUX UZXMN VC2 VFIZW W23 W48 WK8 YLTOR Z45 ZMTXR ~8M ~EX AAYXX AFFHD CITATION
ID	FETCH-LOGICAL-c273w-6aef13aecd9f4e542b2cb41ae208b29329e9bfa7e24aec0c47d660340e34ecc93
IEDL.DBID	RSV
ISSN	0941-0643
IngestDate	Thu Nov 06 12:28:44 EST 2025 Sat Nov 29 04:30:45 EST 2025 Mon Jul 21 06:08:37 EDT 2025
IsDoiOpenAccess	true
IsOpenAccess	true
IsPeerReviewed	true
IsScholarly	true
Issue	4
Keywords	Model partitioning Cosine similarity Vision transformer Cloud computing Token clustering Edge computing
Language	English
LinkModel	DirectLink
MergedId	FETCHMERGED-LOGICAL-c273w-6aef13aecd9f4e542b2cb41ae208b29329e9bfa7e24aec0c47d660340e34ecc93
Notes	ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 14
ORCID	0000-0003-4351-6923
OpenAccessLink	https://link.springer.com/10.1007/s00521-024-10718-w
PQID	3163602178
PQPubID	2043988
PageCount	13
ParticipantIDs	proquest_journals_3163602178 crossref_primary_10_1007_s00521_024_10718_w springer_journals_10_1007_s00521_024_10718_w
PublicationCentury	2000
PublicationDate	20250200
PublicationDateYYYYMMDD	2025-02-01
PublicationDate_xml	– month: 2 year: 2025 text: 20250200
PublicationDecade	2020
PublicationPlace	London
PublicationPlace_xml	– name: London – name: Heidelberg
PublicationTitle	Neural computing & applications
PublicationTitleAbbrev	Neural Comput & Applic
PublicationYear	2025
Publisher	Springer London Springer Nature B.V
Publisher_xml	– name: Springer London – name: Springer Nature B.V
References	E Schubert (10718_CR27) 2021 Z Li (10718_CR9) 2024 K Cao (10718_CR10) 2020; 8 L Alzubaidi (10718_CR1) 2021; 8 10718_CR19 Z Zhao (10718_CR38) 2018; 37 J Chen (10718_CR11) 2019; 107 J Ren (10718_CR6) 2023; 23 10718_CR17 Q Li (10718_CR2) 2024; 21 10718_CR15 10718_CR37 10718_CR14 Y Chen (10718_CR40) 2020; 6 10718_CR35 Z Li (10718_CR33) 2023; 1 SA Molavi (10718_CR20) 2024 10718_CR31 L Zeng (10718_CR18) 2021; 29 L Wang (10718_CR3) 2024 WA Wulf (10718_CR39) 1995; 23 R Agrawal (10718_CR28) 1993 A Joshi (10718_CR8) 2024 CR Harris (10718_CR47) 2020; 585 RAF Belfor (10718_CR26) 1994; 3 10718_CR4 O Russakovsky (10718_CR42) 2015; 115 H Kaneko (10718_CR7) 2023; 6 10718_CR29 W Xu (10718_CR36) 2023; 17 K Han (10718_CR24) 2023; 45 10718_CR48 10718_CR25 10718_CR46 10718_CR23 10718_CR45 X Liu (10718_CR13) 2023; 72 10718_CR22 10718_CR44 10718_CR21 10718_CR43 N Karunanayake (10718_CR5) 2024 10718_CR41 A Bakhtiarnia (10718_CR12) 2022; 153 E Kilcioglu (10718_CR16) 2021; 9 M Kirişci (10718_CR30) 2023; 65 L Zheng (10718_CR32) 2021; 68 Y Kang (10718_CR34) 2017; 45
References_xml	– volume: 8 start-page: 1 year: 2021 ident: 10718_CR1 publication-title: J Big Data doi: 10.1186/s40537-021-00444-8 – volume: 23 start-page: 20 issue: 1 year: 1995 ident: 10718_CR39 publication-title: ACM SIGARCH Comput Arch News doi: 10.1145/216585.216588 – ident: 10718_CR43 – year: 2024 ident: 10718_CR5 publication-title: Neural Comput Appl doi: 10.1007/s00521-024-09910-9 – ident: 10718_CR45 – volume: 3 start-page: 492 issue: 5 year: 1994 ident: 10718_CR26 publication-title: IEEE Trans Image Process doi: 10.1109/83.334991 – volume: 6 start-page: 4345 issue: 5 year: 2023 ident: 10718_CR7 publication-title: Heritage doi: 10.3390/heritage6050230 – volume: 153 start-page: 461 year: 2022 ident: 10718_CR12 publication-title: Neural Netw doi: 10.1016/j.neunet.2022.06.038 – ident: 10718_CR35 doi: 10.23919/DATE.2017.7927211 – ident: 10718_CR37 doi: 10.1109/ICDCS.2017.226 – volume: 9 start-page: 79611 year: 2021 ident: 10718_CR16 publication-title: IEEE Access doi: 10.1109/ACCESS.2021.3084689 – volume: 45 start-page: 615 issue: 1 year: 2017 ident: 10718_CR34 publication-title: ACM SIGARCH Comput Arch News doi: 10.1145/3093337.3037698 – volume: 115 start-page: 211 issue: 3 year: 2015 ident: 10718_CR42 publication-title: Int J Comput Vis (IJCV) doi: 10.1007/s11263-015-0816-y – volume: 68 start-page: 5990 issue: 7 year: 2021 ident: 10718_CR32 publication-title: IEEE Trans Industr Electron doi: 10.1109/TIE.2020.2998756 – volume: 72 start-page: 1 year: 2023 ident: 10718_CR13 publication-title: IEEE Trans Instrum Meas doi: 10.1109/TIM.2023.3264039 – volume: 8 start-page: 85714 year: 2020 ident: 10718_CR10 publication-title: IEEE Access doi: 10.1109/ACCESS.2020.2991734 – volume: 29 start-page: 595 issue: 2 year: 2021 ident: 10718_CR18 publication-title: IEEE/ACM Trans Netw doi: 10.1109/TNET.2020.3042320 – volume: 1 start-page: 15 year: 2023 ident: 10718_CR33 publication-title: IEEE Trans Neural Netw Learn Syst doi: 10.1109/TNNLS.2023.3240195 – ident: 10718_CR48 doi: 10.1145/3373087.3375887 – ident: 10718_CR17 doi: 10.1109/HPCA47549.2020.00036 – year: 2024 ident: 10718_CR20 publication-title: Neural Comput Appl doi: 10.1007/s00521-024-10015-6 – ident: 10718_CR29 – ident: 10718_CR31 doi: 10.1007/978-3-030-69538-5_4 – volume: 107 start-page: 1655 issue: 8 year: 2019 ident: 10718_CR11 publication-title: Proc IEEE doi: 10.1109/JPROC.2019.2921977 – start-page: 32 volume-title: Similarity search and applications year: 2021 ident: 10718_CR27 doi: 10.1007/978-3-030-89657-7_3 – start-page: 69 volume-title: Foundations of data organization and algorithms year: 1993 ident: 10718_CR28 doi: 10.1007/3-540-57301-1_5 – volume: 17 start-page: 9 issue: 1 year: 2023 ident: 10718_CR36 publication-title: IEEE J Sel Top Signal Process doi: 10.1109/JSTSP.2023.3239189 – volume: 6 start-page: 264 issue: 3 year: 2020 ident: 10718_CR40 publication-title: Engineering doi: 10.1016/j.eng.2020.01.007 – ident: 10718_CR4 doi: 10.1016/j.iot.2023.101047 – year: 2024 ident: 10718_CR8 publication-title: Neural Comput Appl doi: 10.1007/s00521-024-09891-9 – ident: 10718_CR21 – ident: 10718_CR46 – ident: 10718_CR19 – ident: 10718_CR44 – ident: 10718_CR15 doi: 10.1145/2809695.2809711 – ident: 10718_CR25 doi: 10.1145/3516807.3516820 – volume: 23 start-page: 6547 issue: 14 year: 2023 ident: 10718_CR6 publication-title: Sensors doi: 10.3390/s23146547 – volume: 585 start-page: 357 issue: 7825 year: 2020 ident: 10718_CR47 publication-title: Nature doi: 10.1038/s41586-020-2649-2 – volume: 65 start-page: 855 issue: 2 year: 2023 ident: 10718_CR30 publication-title: Knowl Inf Syst doi: 10.1007/s10115-022-01776-4 – volume: 37 start-page: 2348 issue: 11 year: 2018 ident: 10718_CR38 publication-title: IEEE Trans Comput Aided Des Integr Circuits Syst doi: 10.1109/TCAD.2018.2858384 – ident: 10718_CR23 doi: 10.1109/ICCV48922.2021.00986 – volume: 45 start-page: 87 issue: 1 year: 2023 ident: 10718_CR24 publication-title: IEEE Trans Pattern Anal Mach Intell doi: 10.1109/TPAMI.2022.3152247 – volume: 21 start-page: 116 issue: 4 year: 2024 ident: 10718_CR2 publication-title: J Real-Time Image Proc doi: 10.1007/s11554-024-01496-8 – ident: 10718_CR22 doi: 10.1109/ICCV48922.2021.00060 – ident: 10718_CR14 doi: 10.1109/INFOCOM.2018.8485905 – year: 2024 ident: 10718_CR3 publication-title: Neural Comput Appl doi: 10.1007/s00521-024-09938-x – year: 2024 ident: 10718_CR9 publication-title: Neural Comput Appl doi: 10.1007/s00521-024-09838-0 – ident: 10718_CR41
SSID	ssj0004685
Score	2.3714094
Snippet	Deploying huge deep learning applications on resource-constrained edge devices is a challenging task. Cloud-based edge computing is a promising solution. Such...
SourceID	proquest crossref springer
SourceType	Aggregation Database Index Database Publisher
StartPage	2627
SubjectTerms	Algorithms Artificial Intelligence Centroids Cloud computing Computational Biology/Bioinformatics Computational Science and Engineering Computer Science Data Mining and Knowledge Discovery Deep learning Edge computing Image Processing and Computer Vision Original Article Partitioning Probability and Statistics in Computer Science Similarity Transmission efficiency Vision Visual tasks
Title	A cosine similarity-based token subsampling method for vision transformer in cloud computing
URI	https://link.springer.com/article/10.1007/s00521-024-10718-w https://www.proquest.com/docview/3163602178
Volume	37
hasFullText	1
inHoldings	1
isFullTextHit
isPrint
journalDatabaseRights	– providerCode: PRVAVX databaseName: SpringerLINK Contemporary 1997-Present customDbUrl: eissn: 1433-3058 dateEnd: 99991231 omitProxy: false ssIdentifier: ssj0004685 issn: 0941-0643 databaseCode: RSV dateStart: 19970101 isFulltext: true titleUrlDefault: https://link.springer.com/search?facet-content-type=%22Journal%22 providerName: Springer Nature
link	http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwnV07T8MwELagMLBQnqJQkAc2iJTYrp2MFaJiqhAvdUCKHNtBEZCgJqV_n7OTtIBggDnOOTqf776L7_whdOqr0E8HSntacuOxiKeeFNpyaED8g5yIa5I4sgkxHoeTSXTdNIWVbbV7eyTpPPWi2c3-wYTUlzBwHeBRvfkqWoNwF1rChpvbh0_dkI6IE_IWW9PDaNMq87OMr-FoiTG_HYu6aDPq_u87t9Bmgy7xsDaHbbRi8h3UbZkbcLORd9HjEKvCVrzjMnvNILkFLO7ZgKZxVTybHJfgT6QtNs-fcE0yjQHd4roTHVct2gWZWY7VSzHTWLlZ4IU9dD-6vLu48hqeBU8BeJl7XJo0oNIoHaXMDBhJiEpYIA3xwwTgAIlMlKRSGMJgkK-Y0Jz7lPmGMrCAiO6jTl7k5gDhhIdchowKwy2PiS9TKmRAha9SgJKJ6KGzVt3xW32dRry4ONkpLgbFxU5x8byH-u2KxM3WKmMa2CvOIJMKe-i8XYHl49-lHf5t-BHaIJbr11Vo91Gnms7MMVpX71VWTk-cyX0AqEfStA
linkProvider	Springer Nature
linkToHtml	http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwnV1NT8MwDLVgIMGF8SkGA3LgBpXaJkvb44SYhoAJwUA7IFVpmqIKaNG6sb-Pk7UbIDjAualTOYn93Nh-AMe29O2kJWMrFlxZLOCJJbxYc2ig_8OYiMduZMgmvF7PHwyCm7IorKiy3asrSWOpZ8Vu-g8mhr4uQ9OBFtWaLMISQ4-lO-bf3j18qoY0RJwYt-icHkbLUpmfZXx1R3OM-e1a1HibTv1_37kOayW6JO3pdtiABZVtQr1ibiDlQd6CxzaRuc54J0X6mmJwi1jc0g4tJqP8WWWkQHsidLJ59kSmJNME0S2ZVqKTUYV2UWaaEfmSj2MizSz4wjbcd877Z12r5FmwJIKXicWFShwqlIyDhKkWcyNXRswRyrX9COGAG6ggSoSnXIaDbMm8mHObMltRhjsgoDtQy_JM7QKJuM-Fz6inuOYxsUVCPeFQz5YJQsnIa8BJpe7wbdpOI5w1TjaKC1FxoVFcOGlAs1qRsDxaRUgd3eIMIym_AafVCswf_y5t72_Dj2Cl27--Cq8uepf7sOpq3l-Trd2E2mg4VgewLN9HaTE8NNvvA5KA1Zg
linkToPdf	http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwpV1LT8MwDLZgIMSF8RSDATlwg4quCWl7nIAJBJom8RAHpCpNUjQB7UQ79vdx0nY8BAfEualTOU78ubH9Aey7MnCTY6kcJbh2WMgTR_jKcGig_8OYiCsvtmQTfr8f3N-Hg09V_Dbbvb6SLGsaTJemtDgaqeRoWvhm_mZiGOwxPEbwdHUmszDHTCK9idev7z5VRlpSToxhTH4Po1XZzM8yvrqmD7z57YrUep5e8__fvAxLFeok3dJMVmBGp6vQrBkdSLXB1-ChS2RmMuFJPnwZYtCLGN0xjk6RInvSKcnxnBEmCT19JCX5NEHUS8oKdVLUKBhlDlMin7OxItLOgi-sw23v7Obk3Kn4FxyJoGbicKGTDhVaqjBh-ph5sSdj1hHac4MYYYIX6jBOhK89hoNcyXzFuUuZqylDywjpBjTSLNWbQGIecBEw6mtu-E1ckVBfdKjvygQhZuy34KBWfTQq22xE04bKVnERKi6yiosmLWjXqxNVWy6PaMe0PsMIK2jBYb0aH49_l7b1t-F7sDA47UVXF_3LbVj0DB2wTeJuQ6N4HesdmJdvxTB_3bWW-A6ZYN58
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=A+cosine+similarity-based+token+subsampling+method+for+vision+transformer+in+cloud+computing&rft.jtitle=Neural+computing+%26+applications&rft.au=Li%2C+Qi&rft.au=Kaneko%2C+Hayata&rft.au=Meng%2C+Lin&rft.date=2025-02-01&rft.issn=0941-0643&rft.eissn=1433-3058&rft.volume=37&rft.issue=4&rft.spage=2627&rft.epage=2639&rft_id=info:doi/10.1007%2Fs00521-024-10718-w&rft.externalDBID=n%2Fa&rft.externalDocID=10_1007_s00521_024_10718_w
thumbnail_l	http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=0941-0643&client=summon
thumbnail_m	http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=0941-0643&client=summon
thumbnail_s	http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=0941-0643&client=summon