Konvolucijske mreže grafova

master's thesis

• Title (english): Graph Convolution Networks
• Author: Natalija Grbac
• Mentor: Lana Horvat Dmitrović (mentor)
• Committee member: Lana Horvat Dmitrović (član povjerenstva)
• Committee member: Mario Krnić (član povjerenstva)
• Committee member: Tomislav Burić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-08, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U okviru ovog rada proučit će se područje primjene konvolucijskih mreža grafova te će se prezentirati i usporediti različiti tipovi istih. Usporedba će se provesti prikladnom programskom implementacijom i primjenom na različitim grafovima socijalnih i drugih mreža koji su javno dostupni. Na početku su objašnjeni osnovni principi rada neuronskih mreža te arhitekture konvolucijskih i povratnih neuronskih mreža. U trećem poglavlju je opisan princip rada neuronskih mreža na grafovima koje slijede rekurzivni pristup polunadziranom učenju na grafovima te su prezentirana dva modela neuronskih mreža grafova . U četvrtom poglavlju je objašnjen princip rada konvolucijskih slojeva na podacima koji nisu iz Euklidske domene. Prikazan je razvoj konvolucijskih mreža grafova koje se oslanjaju na spektralnom pristupu konvoluciji na grafovima, te su detaljno opisane operacije konvolucije i sažimanja u tim mrežama. U zadnjem poglavlju su prezentirani rezultati usporedbe performansa opisanih modela konvolucijskih mreža grafova na zadacima klasifikacije na razini čvorova i na razini grafa.

Abstract (english)

The purpose of this paper is to describe various graph convolutional network models and to compare their performance on both node and graph-level classification tasks on different graphs of social and other networks that are publicly available. The second chapter of this paper describes basic principles of neural networks followed by a detailed explanation of convolutional and recurrent neural network architectures. A recursive approach to semi-supervised learning on graphs is explained in chapter 3 titled Graph neural networks. In this chapter two graph neural network models are presented in detail. The fourth chapter explains the principle of the convolution operation on non-Euclidean domains. The development of graph convolutional networks based on the spectral approach to graph convolution is described. For the graph convolutional models presented in this chapter, the convolutional and pooling operations are described. Finally, the classification performance results for the previously described graph convolutional networks are presented.

• Keywords: onvolucijske mreže grafova; neuronske mreže grafova; spektralna konvolucija; klasifikacija na razini čvorova; klasifikacija na razini grafa
• Keywords (english): graph convolutional networks; graph neural networks; spectral convolution; node-level classification; graph-level classification
• Language: croatian
• URN:NBN: urn:nbn:hr:168:600153
• 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 18:12:44