prikaz prve stranice dokumenta Predstavljanje slika ograničenim Boltzmannovim neuronskim mrežama
No public access
undergraduate thesis
Predstavljanje slika ograničenim Boltzmannovim neuronskim mrežama
Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing, 2015. urn:nbn:hr:168:893042
Relić, Ivan
University of Zagreb
Faculty of Electrical Engineering and Computing

Institutional repository: FER Repository

Cite this document

Relić, I. (2015). Predstavljanje slika ograničenim Boltzmannovim neuronskim mrežama (Undergraduate thesis). Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing. Retrieved from https://urn.nsk.hr/urn:nbn:hr:168:893042

Relić, Ivan. "Predstavljanje slika ograničenim Boltzmannovim neuronskim mrežama." Undergraduate thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, 2015. https://urn.nsk.hr/urn:nbn:hr:168:893042

Relić, Ivan. "Predstavljanje slika ograničenim Boltzmannovim neuronskim mrežama." Undergraduate thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, 2015. https://urn.nsk.hr/urn:nbn:hr:168:893042

Relić, I. (2015). 'Predstavljanje slika ograničenim Boltzmannovim neuronskim mrežama', Undergraduate thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, accessed 22 November 2024, https://urn.nsk.hr/urn:nbn:hr:168:893042

Relić I. Predstavljanje slika ograničenim Boltzmannovim neuronskim mrežama [Undergraduate thesis]. Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing; 2015 [cited 2024 November 22] Available at: https://urn.nsk.hr/urn:nbn:hr:168:893042

I. Relić, "Predstavljanje slika ograničenim Boltzmannovim neuronskim mrežama", Undergraduate thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, 2015. Available at: https://urn.nsk.hr/urn:nbn:hr:168:893042

Metadata
TitlePredstavljanje slika ograničenim Boltzmannovim neuronskim mrežama
Title (english)Representing images with restricted Boltzmann networks
AuthorIvan Relić
MentorSiniša Šegvić (mentor)
Committee memberSiniša Šegvić (predsjednik povjerenstva)
Committee memberMladen Karan (član povjerenstva)
Committee memberGoran Glavaš (član povjerenstva)
GranterUniversity of Zagreb
Faculty of Electrical Engineering and Computing
Zagreb
Defense date and country2015-07-13, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Abstract
Rad opisuje postupak učenja modela koji omogućuje predstavljanje slika u nižoj dimenzionalnosti – kompresiju slika. Kao skup slika koji se predstavlja u nižoj dimenzionalnosti koristi se MNIST skup podataka rukom pisanih brojeva. MNIST skup podataka službeno se sastoji od dva disjunktna skupa – skupa za učenje koji koristimo za učenje same mreže i skupa za ispitivanje koji koristimo za provjeru preciznosti naučenog modela. Skup podataka se prvo normalizira tako da ima vrijednosti između 0 i 1, uključivo. Model koji učimo je ograničena Boltzmannova neuronska mreža. Model učimo algoritmom kontrastne divergencije. Učenje modela takvim algoritmom prikazano je i nad vrlo jednostavnim skupom podataka – zašumljenim podacima s pravca. Eksperimentalni rezultati preciznosti modela nad MNIST skupom podataka smanjene veličine prikazani su koristeći različite parametre algoritma za učenje. Najbolji dobiveni parametri korišteni su za učenje nad cjelokupnim skupom podataka. Dobivena naučena mreža je potom ispitana na testnom skupu podataka uz prikaz najboljih i najgorih rekonstrukcija, prosječnu pogrešku te usporedbu s metodom PCA.
Abstract (english)
This paper describes training of model that enables mapping of original images into lowered dimensionality ones - compressing. As dataset that is represented in lowered dimensionality is used MNIST handwritten digits dataset. MNIST dataset is officially divided into two disjunctive sets – training set which is used for training the model and testing set which is used to check the accuracy of trained model. First thing to do is normalizing the dataset so it contains values from 0 to 1, inclusive. The model that is trained is restricted Boltzmann network. Algorithm used for training is Contrastive Divergence algorithm. Model training with mentioned algorithm is also shown on very simple dataset – noisy line data. Experimental results of model accuracy on reduced size MNIST dataset were shown using different sets of algorithm parameters. The best parameters were used to train the model on complete MNIST dataset. Trained model was then tested on test dataset showing best and worst reconstruction results, average reconstruction error and comparation to PCA method.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:168:893042
Study programmeTitle: Computing
Study programme type: university
Study level: undergraduate
Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
Type of resourceText
File originBorn digital
Access conditionsClosed access
Terms of useLicence In copyright icon
Public note
Created on2018-02-06 11:34:12