AMBIENTUM BIOETHICA BIOLOGIA CHEMIA DIGITALIA DRAMATICA EDUCATIO ARTIS GYMNAST. ENGINEERING EPHEMERIDES EUROPAEA GEOGRAPHIA GEOLOGIA HISTORIA HISTORIA ARTIUM INFORMATICA IURISPRUDENTIA MATHEMATICA MUSICA NEGOTIA OECONOMICA PHILOLOGIA PHILOSOPHIA PHYSICA POLITICA PSYCHOLOGIA-PAEDAGOGIA SOCIOLOGIA THEOLOGIA CATHOLICA THEOLOGIA CATHOLICA LATIN THEOLOGIA GR.-CATH. VARAD THEOLOGIA ORTHODOXA THEOLOGIA REF. TRANSYLVAN
The STUDIA UNIVERSITATIS BABEŞ-BOLYAI issue article summary
The summary of the selected article appears at the bottom of the page. In order to get back to the contents of the issue this article belongs to you have to access the link from the title. In order to see all the articles of the archive which have as author/co-author one of the authors mentioned below, you have to access the link from the author's name.
Authors: TIDOR-VLAD PRICOPE.
DOI: 10.24193/subbi.2021.1.01
Published Online: 2021-06-30
Published Print: 2021-06-30
pp. 5-22
VIEW PDF
FULL PDF
Neural Networks have become a powerful tool in computer vision because of the recent breakthroughs in computation time and model architecture. Very deep models allow for better deciphering of the hidden patterns in the data; however, training them successfully is not a trivial problem, because of the notorious vanishing/exploding gradient problem. We illustrate this problem on VGG models, with 8 and 38 hidden layers, on the CIFAR100 image dataset, where we visualize how the gradients evolve during training. We explore known solutions to this problem like Batch Normalization (BatchNorm) or Residual Networks (ResNets), explaining the theory behind them. Our experiments show that the deeper model suffers from the vanishing gradient problem, but BatchNorm and ResNets do solve it. The employed solutions slighly improve the performance of shallower models as well, yet, the fixed deeper models outperform them.
Keywords and phrases: Deep Learning, Neural Network, Image Classification, Deep Convolutional Neural Network, Vanishing Gradient Problem, VGG.
2010 Mathematics Subject Classification: 68T45.
1998 CR Categories and Descriptors. I.2.1 [Artificial Intelligence]: Learning – Connectionism and neural nets.
