Kombinatorički dizajni za duboko učenje

master's thesis

• Title (english): Combinatorial designs for deep learning
• Author: Ante Bilić
• Mentor: Mario Osvin Pavčević (mentor)
• Committee member: Mario Osvin Pavčević (predsjednik povjerenstva)
• Committee member: Domagoj Kovačević (član povjerenstva)
• Committee member: Mate Puljiz (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-09, Croatia

Abstract

U sklopu ovog diplomskog rada razvijena su dva načina generiranja dropout dizajna. Prvi način je generiranje dropout dizajna genetskim algoritmom, dok je drugi način trivijalno generiranje određenog tipa dropout dizajna. Generirani dropout dizajni su se koristili prilikom treniranja potpuno povezanih konvolucijskih i povratnih neuronskih mreža uz pomoć izrađene implementacije za primjenu generiranih dizajna. Dobiveni rezultati kombinatoričkog dropouta na različitim mrežama su uspoređeni sa... More stohastičkim dropoutom i normalizacijom nad grupom. Korištenjem statističkih testova uspoređeni su rezultati i dobiveni zaključci. Less

Abstract (english)

Within this master ́s thesis, two different ways of generating dropout designs have been developed. The first one is generating dropout designs by genetic algorithm, while the other one is the trivial generating of a specific type of dropout designs. The generated dropout designs were used during the training process of feedforward, convolutional and recurrent neural networks, assisted by custom implementation for using generated designs. The received results of combinatorial dropout on... More different networks have been compared to stochastic dropout and batch normalization. By using statistical tests, the results have been compared and the conclusions have been made. Less

• Keywords: dropout; dropout dizajni; neuronske mreže; normalizacija nad grupom
• Keywords (english): dropout; dropout designs; neural networks; batch normalization
• Language: croatian
• URN:NBN: urn:nbn:hr:168:997343
• 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: Open access
• Created on: 2024-10-12 20:06:33