prikaz prve stranice dokumenta Regularizacija u dubinskom strojnom učenju
Access restricted to students and staff of home institution
master's thesis
Regularizacija u dubinskom strojnom učenju
2019. urn:nbn:hr:166:515162
Marin, Ivana
University of Split
Faculty of Science
Department of Mathematics

Cite this document

Marin, I. (2019). Regularizacija u dubinskom strojnom učenju (Master's thesis). Split: University of Split, Faculty of Science. Retrieved from https://urn.nsk.hr/urn:nbn:hr:166:515162

Marin, Ivana. "Regularizacija u dubinskom strojnom učenju." Master's thesis, University of Split, Faculty of Science, 2019. https://urn.nsk.hr/urn:nbn:hr:166:515162

Marin, Ivana. "Regularizacija u dubinskom strojnom učenju." Master's thesis, University of Split, Faculty of Science, 2019. https://urn.nsk.hr/urn:nbn:hr:166:515162

Marin, I. (2019). 'Regularizacija u dubinskom strojnom učenju', Master's thesis, University of Split, Faculty of Science, accessed 27 December 2024, https://urn.nsk.hr/urn:nbn:hr:166:515162

Marin I. Regularizacija u dubinskom strojnom učenju [Master's thesis]. Split: University of Split, Faculty of Science; 2019 [cited 2024 December 27] Available at: https://urn.nsk.hr/urn:nbn:hr:166:515162

I. Marin, "Regularizacija u dubinskom strojnom učenju", Master's thesis, University of Split, Faculty of Science, Split, 2019. Available at: https://urn.nsk.hr/urn:nbn:hr:166:515162

Please login to the repository to save this object to your list.
Metadata
TitleRegularizacija u dubinskom strojnom učenju
Title (english)Regularization in Deep Learning
AuthorIvana Marin
MentorŽeljko Agić (mentor)
Committee memberŽeljko Agić (predsjednik povjerenstva)
Committee memberJurica Perić (član povjerenstva)
Committee memberGordan Radobolja (član povjerenstva)
GranterUniversity of Split
Faculty of Science
(Department of Mathematics)
Split
Defense date and country2019-09-26, Croatia
Scientific / art field,
discipline and subdiscipline		NATURAL SCIENCES
Mathematics
Abstract
Jedan od glavnih izazova treniranja dubokih neuronskih mreža s milijunima parametara je izbjegavanje prenaučenosti koja rezultira lošom generalizacijom na novim podatcima. Problemu prenaučenosti može se pristupiti dodavanjem novih podataka skupu za treniranje ili regularizacijom. Cilj ovog diplomskog rada je dati pregled različitih metoda regularizacije koje se koriste u dubinskom strojnom učenju te eksperimentalno ispitati njihov utjecaj na generalizacijsku pogrešku modela. Eksperimentima je pokazano da se odabirom prikladnih metoda regularizacije za dani problem performanse modela na novim podatcima mogu znatno poboljšati.
Abstract (english)
One of the major challenges of training deep neural networks with millions of parameters is avoiding overfitting which results in bad generalization on new data. There are two main approaches to problem of overfitting, collecting more training data and regularization. The main objective of this master’s thesis is to give an overview of different regularization methods and experimentally examine their effect on generalization error. Experiments showed that model performance on new data can be greately improved by choosing appropriate regularization techniques for a given problem.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:166:515162
Study programmeTitle: Mathematics; specializations in: major in pure mathematics, major in computing, major in teaching
Course: major in computing
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra matematike (magistar/magistra matematike)
Type of resourceText
File originBorn digital
Access conditionsAccess restricted to students and staff of home institution
Terms of useLicence In copyright icon
Created on2020-06-08 09:33:42