COMPARISON OF WELL-KNOWN CONVOLUTIONAL NEURAL NETWORKS IMAGE CLASSIFICATION MODELS ON THE CIFAR-10 DATASET / UMAR BINDIR AHMED; SUPERVISOR: ASSOC. PROF. DR. UMAR ÖZGÜNALP

Yazar: Katkıda bulunan(lar):Dil: İngilizce 2023Tanım: xi, 62 sheets; 31 cm. 1 CD-ROMİçerik türü:
  • text
Ortam türü:
  • unmediated
Taşıyıcı türü:
  • volume
Konu(lar): Tez notu: Thesis (MSc) - Cyprus International University. Institute of Graduate Studies and Research Electrical and Electronics Engineering Department Özet: ABSTRACT In this research, a supervised learning method is used to assess the performance of well-known deep convolutional neural network models for image classification on the CIFAR-10 dataset. VGGNet, GoogLeNet, ResNet, and AlexNet are the four models that were trained and evaluated using a variety of measures, including loss and accuracy. ResNet and VGGNet obtained the greatest test accuracy of 78.15% when comparing the performances of deep convolutional neural network models during training and testing. AlexNet and GoogleNet came in second and third, with the highest test accuracy of 77.5% and 74.41%, respectively. These outcomes suggest that the deep convolutional neural network models in a supervised method are capable of successfully categorizing CIFAR-10 images with reasonably small loss values, as shown in Table 4.1. The findings underscore the significance of meticulous planning and execution of training and testing methodologies for deep convolutional neural networks in supervised learning scenarios to efficiently exploit data with annotations. One may investigate various strategies, including data augmentation, regularization, and learning rate scheduling, to enhance the effectiveness of the models. The result of this investigation emphasizes the importance of precise preparation and implementation of training and testing procedures in deep convolutional neural networks in supervised learning in order to maximize their capacity for learning. Further investigation is necessary to tackle the limitations and challenges that occur with maximizing the use of data with annotations. The study at issue offers major findings regarding the impact and implications of the models utilized on the CIFAR-10 dataset. The findings contribute to an understanding of the strengths and weaknesses of these approaches, guiding future research to enhance the performance of the image classification models. Keywords: Alexnet, CIFAR-10 Dataset, Convolutional Neural Networks, Googlenet, Image Classification, Resnet, Vggnet
Materyal türü: Thesis
Mevcut
Materyal türü Geçerli Kütüphane Koleksiyon Yer Numarası Durum Notlar İade tarihi Barkod Materyal Ayırtmaları
Thesis Thesis CIU LIBRARY Tez Koleksiyonu Tez Koleksiyonu YL 3165 A46 2023 (Rafa gözat(Aşağıda açılır)) Kullanılabilir Electrical and Electronics Engineering Department T3546
Suppl. CD Suppl. CD CIU LIBRARY Görsel İşitsel YL 3165 A46 2023 (Rafa gözat(Aşağıda açılır)) Kullanılabilir Electrical and Electronics Engineering Department CDT3546
Toplam ayırtılanlar: 0

Thesis (MSc) - Cyprus International University. Institute of Graduate Studies and Research Electrical and Electronics Engineering Department

Includes bibliography (sheets 60-62)

ABSTRACT
In this research, a supervised learning method is used to assess the performance of
well-known deep convolutional neural network models for image classification on the
CIFAR-10 dataset. VGGNet, GoogLeNet, ResNet, and AlexNet are the four models
that were trained and evaluated using a variety of measures, including loss and
accuracy.
ResNet and VGGNet obtained the greatest test accuracy of 78.15% when comparing
the performances of deep convolutional neural network models during training and
testing. AlexNet and GoogleNet came in second and third, with the highest test
accuracy of 77.5% and 74.41%, respectively. These outcomes suggest that the deep
convolutional neural network models in a supervised method are capable of
successfully categorizing CIFAR-10 images with reasonably small loss values, as
shown in Table 4.1.
The findings underscore the significance of meticulous planning and execution of
training and testing methodologies for deep convolutional neural networks in
supervised learning scenarios to efficiently exploit data with annotations. One may
investigate various strategies, including data augmentation, regularization, and
learning rate scheduling, to enhance the effectiveness of the models.
The result of this investigation emphasizes the importance of precise preparation and
implementation of training and testing procedures in deep convolutional neural
networks in supervised learning in order to maximize their capacity for learning.
Further investigation is necessary to tackle the limitations and challenges that occur
with maximizing the use of data with annotations.
The study at issue offers major findings regarding the impact and implications of the
models utilized on the CIFAR-10 dataset. The findings contribute to an understanding of the strengths and weaknesses of these approaches, guiding future research to enhance the performance of the image classification models.
Keywords: Alexnet, CIFAR-10 Dataset, Convolutional Neural Networks, Googlenet,
Image Classification, Resnet, Vggnet

Araştırmaya Başlarken  
   Sıkça Sorulan Sorular