Procjenjivanje nesigurnosti predikcija diskriminativnih konvolucijskih modela primjenom suparničkog učenja

master's thesis

• Title (english): Estimating prediction uncertainty of discriminative convolutional models with adversarial learning
• Author: Josip Šarić
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Jan Šnajder (član povjerenstva)
• Committee member: Tomislav Hrkać (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2018-07-19, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Procjena nesigurnosti predikcija važna je osobina sustava dubokog učenja. U ovome radu dan je pregled postojećih postupaka za procjenu nesigurnosti i predložen je takav postupak zasnovan na suparničkom učenju. Implementiran je model suparničkog učenja kod kojeg je generator model semantičke segmentacije zasnovan na arhitekturi ResNet-50, a diskriminator na svome izlazu daje gustu mapu procjene nesigurnosti predikcija generatora. Pokazan je utjecaj suparničkog gubitka na performanse modela u zadatku semantičke segmentacije. Uspoređene su i evaluirane procjene nesigurnosti funkcijom softmax i kao izlaz diskriminatora u suparničkom učenju.

Abstract (english)

Estimating prediction uncertainty is important component of deep learning systems. In this thesis, an overview of existing methods for uncertainty estimation is given and one based on adversarial learning is proposed. Adversarial learning model is implemented in which generator is semantic segmentation model based on ResNet-50 architecture and discriminators output is uncertainty estimation map at pixel level of generators predictions. An impact of adversarial loss on performances of semantic segmentation model was shown. Uncertainty estimation using softmax function and discriminators output in adversarial learning were compared and evaluated.

• Keywords: duboko učenje; semantička segmentacija; procjena nesigurnosti; duboki konvolucijski modeli; suparničko učenje
• Keywords (english): deep learning; semantic segmentation; uncertainty estimation; deep convolutional models; adversarial learning
• Language: croatian
• URN:NBN: urn:nbn:hr:168:224283
• 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: 2019-03-14 01:08:44