Izlučivanje slikovnih reprezentacija generativnim suparničkim modelima

master's thesis

• Title (english): Extracting image representations with generative adversarial models
• Author: Filip Zelić
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Zoran Kalafatić (član povjerenstva)
• Committee member: Tomislav Hrkać (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2018-02-20, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Nenadzirano učenje je vrlo zanimljivo područje jer omogućuje iskoristiti velike količine dostupnih neoznačenih podataka. U radu se bavimo generativnim suparničkim modelima koji omogućuju generiranje slika iz distribucije podatke te izvlačenje slikovnih reprezentacija. Opisane su duboke neuronske mreže s fokusom na konvolucijske mreže. Detaljno je prikazano nekoliko tipova generativnih suparničkih modela i njihove konfiguracije. Implementirani su duboki generativni suparnički model i dvije inačice Wassersteinovog modela u Pytorch razvojnom okviru. Modeli se evaluirani na tri različita podatkovna skupa MNIST, Fashion-MNIST i CIFAR-10. Na kraju su prikazani dobiveni rezultati, s opisima i slikama.

Abstract (english)

Unsupervised learning is a very interesting area because it allows one to take advantage of large volumes of unlabeled data. This paper deals with generative adversarial networks that enable generation of images from given data distribution and feature extraction. Deep neural networks are described with emphasis on convolutional networks. Several types of generative adversarial networks are shown with their configuration. A deep convolutional adversarial model and two versions of Wasserstein model are implemented using the Pytorch library. Models were evaluated on three different sets of data MNIST, Fashion-MNIST and CIFAR-10. Finally, we present results with description and images.

• Keywords: Duboki generativni suparnički modeli; učenje reprezentacije; nenadzirano učenje
• Keywords (english): Deep generative adversarial models; representation learning; unsupervised learning
• Language: croatian
• URN:NBN: urn:nbn:hr:168:662749
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2019-03-14 02:18:29