Institutional Repository of Vilnius University
Konvoliucinių neuroninių tinklų paieška nuotraukoms klasifikuoti /
Abstract
dc:descriptionThe paper compares two algorithms: NAO and CNAS. Both algorithms search for convolutional neural network architectures. Comparison is made by evaluating the found architecture on a dataset and then deriving an F1-score from classification metrics while also considering time. In the paper, the comparison is made by changing packet size and epoch parameters, using additional tools: auxiliary classificator and image cutouts, and different datasets. The results show that the NAO algorithm performed better using the CIFAR10 dataset with cutouts, auxiliary classificator, and likely with smaller packet sizes and more epochs. However, the CNAS algorithm was superior when using the CIFAR10 dataset with a larger packet size and the Tiny ImageNet dataset. Also, the findings suggest that CNAS is more adaptive to changes in the number of classes and the number of images per class. In other cases, both algorithms performed similarly.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Karlonas, Paulius,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD146236974&prefLang=en_US