Sustav za generiranje teksta temeljen na neuronskim mrežama

master's thesis

• Title (english): A Neural-Based Text Generation System
• Author: Tin Kuculo
• Mentor: Jan Šnajder (mentor)
• Committee member: Jan Šnajder (predsjednik povjerenstva)
• Committee member: Domagoj Jakobović (član povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2018-09-26, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U ovom radu istražuju se novije arhitekture umjetnih neuronskih mreža te se testiraju na zadatku generiranja teksta, uz fokus na neuronskim mrežama s proširenim pamćenjem.
Generirani tekst uspoređuje se s trenutačno često korištenim dubokim modelima poput LSTM mreža, ponavljajućim mrežama s dugim kratkoročnim pamćenjem. Razmatraju se tri vrste novijih arhitektura: neuronski Turingovi strojevi, diferencijabilna neuronska računala i duboke Q-mreže.

Abstract (english)

This work explores newer artificial neural network architectures and tests them on the task of text generation, with a focus on memory augmented neural networks. Generated texts are compared with currently often-used deep models like LSTM networks–recurrent networks with long short-term memory. Three new types of architectures are observed: neural Turing machines, differentiable neural computers, and deep Q-networks.

• Keywords: Duboko učenje; generiranje teksta; LSTM; neuronski turing-strojevi; diferencijabilna neuronska računala; duboke Q-mreže
• Keywords (english): Deep learning; text generation; LSTM; neural turing machines; differentiable neural computers; deep Q-networks
• Language: english
• URN:NBN: urn:nbn:hr:168:831282
• 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: 2019-03-13 22:03:26