A novel similarity based unsupervised technique for training convolutional filters

Basit Öğe Kaydı
dc.authorid0000-0001-9033-8934
dc.authorid0000-0003-0298-0690
dc.contributor.authorErkoç, Tuğbaen_US
dc.contributor.authorEskil, Mustata Taneren_US
dc.date.accessioned2023-06-05T07:30:38Z
dc.date.available2023-06-05T07:30:38Z
dc.date.issued2023-05-17
dc.departmentIşık Üniversitesi, Mühendislik ve Doğa Bilimleri Fakültesi, Bilgisayar Mühendisliği Bölümüen_US
dc.departmentIşık University, Faculty of Engineering and Natural Sciences, Department of Computer Engineeringen_US
dc.departmentIşık University, Pattern Recognition and Machine Intelligence Laboratoryen_US
dc.descriptionThis work was supported by the Scientific and Technological Research Council of Turkey (TUBITAK) under Grant 118E293.en_US
dc.description.abstractAchieving satisfactory results with Convolutional Neural Networks (CNNs) depends on how effectively the filters are trained. Conventionally, an appropriate number of filters is carefully selected, the filters are initialized with a proper initialization method and trained with backpropagation over several epochs. This training scheme requires a large labeled dataset, which is costly and time-consuming to obtain. In this study, we propose an unsupervised approach that extracts convolutional filters from a given dataset in a self-organized manner by processing the training set only once without using backpropagation training. The proposed method allows for the extraction of filters from a given dataset in the absence of labels. In contrast to previous studies, we no longer need to select the best number of filters and a suitable filter weight initialization scheme. Applying this method to the MNIST, EMNIST-Digits, Kuzushiji-MNIST, and Fashion-MNIST datasets yields high test performances of 99.19%, 99.39%, 95.03%, and 90.11%, respectively, without applying backpropagation training or using any preprocessed and augmented data.en_US
dc.description.sponsorshipTürkiye Bilimsel ve Teknolojik Araştırma Kurumu (TÜBİTAK)en_US
dc.description.versionPublisher's Versionen_US
dc.identifier.citationErkoç, T. & Eskil, M. T. (2023). A novel similarity based unsupervised technique for training convolutional filters. IEEE Access, 11, 49393-49408. doi:10.1109/ACCESS.2023.3277253en_US
dc.identifier.doi10.1109/ACCESS.2023.3277253
dc.identifier.endpage49408
dc.identifier.issn2169-3536
dc.identifier.scopus2-s2.0-85160234955
dc.identifier.scopusqualityQ1
dc.identifier.startpage49393
dc.identifier.urihttps://hdl.handle.net/11729/5566
dc.identifier.urihttp://dx.doi.org/10.1109/ACCESS.2023.3277253
dc.identifier.volume11
dc.identifier.wosWOS:001006425600001
dc.identifier.wosqualityQ2
dc.indekslendigikaynakWeb of Scienceen_US
dc.indekslendigikaynakScopusen_US
dc.indekslendigikaynakScience Citation Index Expanded (SCI-EXPANDED)en_US
dc.institutionauthorErkoç, Tuğbaen_US
dc.institutionauthorEskil, Mustata Taneren_US
dc.institutionauthorid0000-0001-9033-8934
dc.institutionauthorid0000-0003-0298-0690
dc.language.isoenen_US
dc.peerreviewedYesen_US
dc.publicationstatusPublisheden_US
dc.publisherIEEEen_US
dc.relation.ispartofIEEE Accessen_US
dc.relation.publicationcategoryMakale - Uluslararası Hakemli Dergi - Kurum Öğretim Elemanıen_US
dc.rightsinfo:eu-repo/semantics/openAccessen_US
dc.subjectBackpropagationen_US
dc.subjectComputer architectureen_US
dc.subjectConvolutional neural networksen_US
dc.subjectFeature extractionen_US
dc.subjectMicroprocessorsen_US
dc.subjectTask analysisen_US
dc.subjectTrainingen_US
dc.subjectUnsupervised learningen_US
dc.subjectExtractionen_US
dc.subjectJob analysisen_US
dc.subjectNeural networksen_US
dc.subjectBackpropagation trainingen_US
dc.subjectInitialization methodsen_US
dc.subjectLabeled dataseten_US
dc.subjectTraining schemesen_US
dc.subjectUnsupervised techniquesen_US
dc.subjectNetworken_US
dc.subjectRepresentationen_US
dc.subjectNeocognitronen_US
dc.titleA novel similarity based unsupervised technique for training convolutional filtersen_US
dc.typeArticleen_US
dspace.entity.typePublication

Dosyalar

Orijinal paket
Listeleniyor 1 - 1 / 1
Küçük Resim
İsim:
A_Novel_Similarity_Based_Unsupervised_Technique_for_Training_Convolutional_Filters.pdf
Boyut:
3.47 MB
Biçim:
Adobe Portable Document Format
Açıklama:
Publisher's Version
İndir
Lisans paketi
Listeleniyor 1 - 1 / 1
Küçük Resim Yok
İsim:
license.txt
Boyut:
1.44 KB
Biçim:
Item-specific license agreed upon to submission
Açıklama:
İndir

Koleksiyon

Makale Koleksiyonu | Bilgisayar Mühendisliği Bölümü
Scopus İndeksli Yayınlar Koleksiyonu
WoS İndeksli Yayınlar Koleksiyonu