Tehnike dubokog učenja za podatke strukturirane u obliku grafova

master's thesis

• Title (english): Deep Learning Techniques on Graph-Structured Data
• Author: Marko Pavlić
• Mentor: Igor Čavrak (mentor)
• Committee member: Igor Čavrak (predsjednik povjerenstva)
• Committee member: Danko Basch (član povjerenstva)
• Committee member: Tomislav Lipić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Graf je sveprisutna struktura podataka na kojoj je moguće izvoditi zadatke dubokog učenja pomoću različitih tehnika. U ovom su radu opisane poznate graf duboke neuronske mreže uz objašnjenje teorijskih osnova i navođenje njihovih prednosti i nedostataka. Također je iznesen i sveobuhvatan pregled područja dubokog učenja na grafovima, a postojeće tehnike dubokog učenja dodatno su objašnjene u kontekstu graf dubokih neuronskih mreža. U sklopu rada implementirani su i robusno evaluirani različiti modeli graf dubokih neuronskih mreža, kako na skupovima podataka koji sadrže grafove citiranja, tako i na skupu podataka koji sadrži detektirane neurone iz slika histopatoloških preparata u cerebralnom korteksu mozga.

Abstract (english)

Graph is a ubiquitous data structure, effective for deep learning tasks using various techniques. This thesis describes well-known graph deep neural networks along with techniques' theoretical bases and their advantages and disadvantages. Furthermore, this thesis gives a comprehensive study of deep learning on graphs. Understanding of existent deep learning techniques is upgraded through the scope of graph deep neural networks. In addition, several techniques are implemented, trained and robustly evaluated on miscellaneous datasets, including citation graphs and neurons detected on images of histopathological slides of brain's cerebral cortex.

• Keywords: duboko učenje; graf; reprezentacija vrha; nadzirano i nenadzirano učenje; graf konvolucijska mreža; induktivnost; mehanizam pažnje; kombinatorna generalizacija; relacijske induktivne sklonosti; Python; PyTorch; PyTorch Geometric
• Keywords (english): deep learning; graph; node embedding; supervised and unsupervised learning; graph convolutional network; inductive learning; attention mechanism; combinatorial generalization; relational inductive biases; Python; PyTorch; PyTorch Geometric
• Language: croatian
• URN:NBN: urn:nbn:hr:168:652454
• 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 21:05:34