prikaz prve stranice dokumenta Generativne kontradikcijske neuronske mreže u sintezi originalnih rješenja
  Download
PDF 2.24 MB
undergraduate thesis
Generativne kontradikcijske neuronske mreže u sintezi originalnih rješenja
Zagreb: University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture, 2021. urn:nbn:hr:235:772937
Homolak, Sandi
University of Zagreb
Faculty of Mechanical Engineering and Naval Architecture

Institutional repository: Repository of Faculty of Mechanical Engineering and Naval Architecture University of Zagreb

Cite this document

Homolak, S. (2021). Generativne kontradikcijske neuronske mreže u sintezi originalnih rješenja (Undergraduate thesis). Zagreb: University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture. Retrieved from https://urn.nsk.hr/urn:nbn:hr:235:772937

Homolak, Sandi. "Generativne kontradikcijske neuronske mreže u sintezi originalnih rješenja." Undergraduate thesis, University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture, 2021. https://urn.nsk.hr/urn:nbn:hr:235:772937

Homolak, Sandi. "Generativne kontradikcijske neuronske mreže u sintezi originalnih rješenja." Undergraduate thesis, University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture, 2021. https://urn.nsk.hr/urn:nbn:hr:235:772937

Homolak, S. (2021). 'Generativne kontradikcijske neuronske mreže u sintezi originalnih rješenja', Undergraduate thesis, University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture, accessed 17 November 2024, https://urn.nsk.hr/urn:nbn:hr:235:772937

Homolak S. Generativne kontradikcijske neuronske mreže u sintezi originalnih rješenja [Undergraduate thesis]. Zagreb: University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture; 2021 [cited 2024 November 17] Available at: https://urn.nsk.hr/urn:nbn:hr:235:772937

S. Homolak, "Generativne kontradikcijske neuronske mreže u sintezi originalnih rješenja", Undergraduate thesis, University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture, Zagreb, 2021. Available at: https://urn.nsk.hr/urn:nbn:hr:235:772937

Metadata
TitleGenerativne kontradikcijske neuronske mreže u sintezi originalnih rješenja
Title (english)Generative adversarial neural networks for synthesis of original solutions
AuthorSandi Homolak
MentorPetar Ćurković (mentor)
GranterUniversity of Zagreb
Faculty of Mechanical Engineering and Naval Architecture
Zagreb
Defense date and country2021, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Artificial Intelligence
Abstract
Generativne kontradikcijske neuronske mreže jedan su od najuspješnijih generativnih modela strojnog učenja. Njihovo izvođenje se temelji na natjecanju dviju neuronskih mreža u hipotetskoj igri s nultom sumom. Za razliku od mnogih drugih generativnih modela, GEKON mreže ne koriste funkcije gustoće vjerojatnosti kako bi opisale distribuciju vjerojatnosti već generiraju nove vrijednosti isključivo na temelju proučavanja značajki na velikim bazama podataka. U ovom radu prikazana je temeljna teorijska i matematička pozadina neuronskih mreža, nakon čega je detaljnije opisana struktura GEKON modela i mogućnost njegove optimizacije za specifične zadatke. U radu su također prikazani primjeri uporabe takvog modela s naglaskom na primjenu u robotici, ali i etička pitanja koja nastaju zbog iznimno realističnih rezultata takvih mreža. Na kraju rada prikazana je i usporedba implementacije triju vrsta GEKON modela za generiranje realističnih „rukom napisanih“ znamenki.
Abstract (english)
Generative adversarial networks are one of the most successful generative models of machine learning. Their performance is based on the competition of two neural networks in a hypothetical zero-sum game. Unlike many other generative models, GANs do not use probability density functions to describe the probability distribution, but rather generate new values solely based on the study of features on large databases. This paper presents the basic theoretical and mathematical background of neural networks, after which the structure of GANs and the possibilities of its optimization for specific tasks are described in more detail. The paper also presents examples of using such a model with an emphasis on the applications in robotics, but also ethical issues that arise due to the extremely realistic results of such networks. A comparison of the implementation of three types of GANs for generating realistic "handwritten" digits is presented at the end of the paper.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:235:772937
Study programmeTitle: Mechanical Engineering
Study programme type: university
Study level: undergraduate
Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka strojarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka strojarstva)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
Created on2021-02-17 00:31:03