Optimiranje strukture neuronske mreže s radijalnim baznim funkcijama primjenom algoritma roja čestica

master's thesis

• Title (english): Particle swarm optimisation of radial basis function neural networks
• Author: Matej Vitasović
• Mentor: Danko Brezak (mentor)
• Granter: University of Zagreb; Faculty of Mechanical Engineering and Naval Architecture; Zagreb
• Defense date and country: 2016-07-14, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Mechanical Engineering

Abstract

U ovom radu je izrađena programska podrška za neuronsku

mrežu s radijalnim baznim funkcijama (RBF mrežu) učenu klasičnom metodom učenja u jednom koraku. Model je proširen osnovnim algoritmom optimizacije rojem čestica (PSO), te s tri njegove modifikacije: unifikacijskim algoritmom optimizacije rojem čestica (UPSO), algoritmom optimizacije rojem čestica s osiguranom konvergencijom (GCPSO), te metodom suprotnih vrijednosti uz korištenje Cauchyevog mutacijskog operatora (OPSO Cauchy). Algoritmi su primijenjeni u optimizaciji vrijednosti elemenata glavne ijagonale kovarijantnih matrica vezanih uz neurone skrivenog sloja RBF mreže. Usporedba svih varijanti algoritama izvršena je testiranjem mreže sa tri klasifikacijska problema: „Krug u kvadratu“, Iris setu podataka, te na određivanju stupnja istrošenosti alata kod bušenja kompozita. Funkcija cilja korištena kod optimizacije bilo je postotak uspješno klasificiranih uzoraka iz seta za testiranje.

Abstract (english)

In this study, a programming support was built for a radial basis function neural network (RBF network) using a classical one-step learning method. The model was expanded with the basic particle swarm optimisation algorithm (PSO), and with three of its modifications: Unified Particle Swarm Optimization (UPSO), Guaranteed Convergence Particle Swarm Optimization (GCPSO), and Opposition-Based Particle Swarm Optimisation with Cauchy Mutation (OPSO Cauchy). The algorithms were used to optimise the values of the elements on the main diagonal of covariance matrices associated with the hidden-layer neurons of the RBF network. A comparison of all algorithms was made by testing the network on three

classification problems: „Circle-in-the-Square“, Iris dataset, and determining the tool wear in drilling composite materials. The objective function used in optimisation methods was the percentage of successfully classified test data samples.

• Keywords: strojno učenje; neuronska mreža; radijalna bazna funkcija; optimizacija rojem čestica; klasifikacija
• Keywords (english): machine learning; neural networks; radial basis function; particle swarm optimization; classification
• Language: croatian
• URN:NBN: urn:nbn:hr:235:983795
• Study programme: Title: Mechanical Engineering; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka strojarstva (magistar/magistra inženjer/inženjerka strojarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2019-05-22 16:47:02