Institutional Repository of Vilnius University
Generatyvinių prešiškų tinklų tyrimas modelį papildant naujais uždaviniais /
Abstract
dc:descriptionIn this paper an overview about generative adversarial networks and few of it’s modifications was made. It was described how traditional generative adversarial network is working and what are the differences between it’s modifcations, such as CGAN, InfoGAN and AC-GAN. Expirements were made with these GAN modifications and 3 low resolution datasets - „fashion-MNIST”, „CIFAR-10”, „CelebA”. During these experiments learning metrics where tracked and intermediate results were stored. The effect of modifications on stability and results was observed during experiments. After experiments the conclusions is that labeled data improved performance of the network sometimes it even learned the dataset faster. But during unsupervised learning while trying InfoGAN network results were worse, training was unstable and results were worse.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Gedgaudas, Deimantas,
- Contributors dc:contributor
-
- Petkevičius, Linas
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:ELABAETD81706952&prefLang=en_US