Vizualizacija i interpretacija konvolucijskih modela

undergraduate thesis

• Title (english): Visualization and Interpretation of Convolutional Models
• Author: Martin Sršen
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Marin Oršić (član povjerenstva)
• Committee member: Josip Šarić (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

Duboki konvolucijski modeli su glavni sastojak mnogih praktičnih primjera računalnog vida. Međutim, često se javlja kritika da takvi modeli nisu spremni za industrijsku upotrebu zbog loše interpretabilnosti, odnosno, nemogućnost modela da svoju odluku obrazloži ljudima. Zbog toga postupci za interpretiranje odluka i vizualiziranje naučenih parametara konvolucijskih modela predstavljaju zanimljivo područje istraživanja.U okviru rada, odabran je okvir za automatsku diferencijaciju te upoznate bibilioteke za rukovanje matricama i slikama. Ukratko su opisani postojeći postupci za interpretiranje i vizualiziranje konvolucijskih modela, njihovih naučenih filtera te načina na koji vide ulazne slike.

Abstract (english)

Deep convolution models are the main ingredient of many practical examples of computer vision. However, there is often criticism that such models are not ready for industrial use due to poor interpretability, that is, the inability of the model to explain its decision to people. Therefore, the procedures for interpretation of decisions and the visualization of the learned parameters of convection models represent an interesting field of research. Within this work, the framework for automatic differentiation and the familiar library for handling matrices and images were selected. The existing procedures for interpreting and visualizing convolutional models, their learned filters, and the way in which they see input images are described briefly.

• Keywords: računalni vid; konvolucijske neuronske mreže; pytorch; strojno učenje; vizualizacija konvolucijske neuronske mreže
• Keywords (english): computer vision; convolutional neural network; pytorch; machine learning; visualization of convolutional neural network
• Language: croatian
• URN:NBN: urn:nbn:hr:168:698247
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-01-07 22:05:50