Usporedba arhitektura generativnih suparničkih mreža za pisanje kratkih tekstualnih poruka

master's thesis

• Title (english): Comparison of Architectures of Generative Adversarial Networks for Writing Short Text Messages
• Author: Ivan Pešorda
• Mentor: Tomislav Petković (mentor)
• Committee member: Tomislav Petković (predsjednik povjerenstva)
• Committee member: Davor Petrinović (član povjerenstva)
• Committee member: Ana Sović Kržić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-12, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Ovaj rad se bavi usporedbom dvaju generativno suparničkih modela za pisanje teksta: SeqGAN i MaskGAN. Rad je prije svega usredotočen na pisanje kratkih poruka pomoću dva gore navedena modela. U uvodnom djelu je objašenjen problem pisanja kratkih poruka, navedeni su slični radovi na ovom području te je dan kratak pregled generativnih suparničkih mreža, SeqGAN mreže i MaskGAN mreže. Nadalje, u radu je objašnjena arhitektura i način ućenja modela korištenih u usporedbi. Nad naučenim modelima je provedeno ljudsko ispitivanje, BLEU metrika, Self-BLEU metrika te je izračunat broj jedinstvenih N-grama. U zaključku rada su komentirani rezultati i postupci učenja.

Abstract (english)

This thesis concerns itself with comparison of two generative adversarial networks for text generation: SeqGAN and MaskGAN. The primary focus of the thesis is generating short messages using the two aforementioned models. Generation of short messages, as well as related work and short explanation of generative adversarial networks, MaskGAN and SeqGAN, are given in the introduction. Furthermore, architecture and training procedure of the models used in comparison are explained in the paper. We performed human evaluation, BLEU metric, Self-BLEU metric and unique N-gram metric and the trained models. Analysis of the results and of the training procedures is given in conclusion.

• Keywords: strojno učenje; generativne suparničke mreže; generiranje teksta; SeqGAN; MaskGAN
• Keywords (english): machine learning; generative adversarial networks; text generation; SeqGAN; MaskGAN
• Language: croatian
• URN:NBN: urn:nbn:hr:168:870513
• 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: 2020-01-07 21:16:34