Učinkovitost optimizacijskih algoritama u ovisnosti o značajkama problema

undergraduate thesis

• Title (english): Optimization efficiency based on fitness landscape
• Author: Vlaho Poluta
• Mentor: Domagoj Jakobović (mentor)
• Committee member: Domagoj Jakobović (predsjednik povjerenstva)
• Committee member: Alan Jović (član povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2014-07-14, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Ovaj rad se bavi ispitivanjem dali se mogu klasificirati optimizacijski problemi i onda odrediti dali je neki algoritam optimizacije dobar za optimiranje određene klase problema, sa ciljem smanjenja vremena potrebnog za određivanjem dobrog optimizacijskog algoritma na novom optimizacijskom problemu. Algoritmi se pokušavaju klasificirati pomoću krajolika dobrote tih problema. Testne funkcije korištene u radu grupirane su u klase pomoću K-means algoritma. U ovom radu korištena su 4 modela klasifikacije strojnim učenjem: stablo odluke generirano genetskim programiranjem, simbolička regresija generirana genetskim programiranjem, učenje neuronske mreže diferencijskom evolucijom i učenje neuronske mreže algoritmom selekcije klona. Ispitivanje se izvodi u 2 faze, u prvoj fazi se određuju najbolji parametri za svaki od modela učenja, dok se u drugoj fazi uspoređuju modeli i procjenjuje se njihova uspješnost klasifikacije.

Abstract (english)

This paper is concerned with examining whether you can classify an optimization problem and then determine is some optimization algorithm good for optimizing that problem with goal of reducing time required for determining that. Algorithms where classified using fitness landscape values. Fitness landscape values were calculated from optimisation problems. Test functions where classified using k-means algorithm. In this paper 4 classification models were used: symbolic regression generated by genetic programming, decision three generated by genetic programming, neural network trained by differential evolution and finaly neural network trained by clonal selection algorithm. Testing was done in 2 faze. In first faze best parameters for all modes where determent, while in the second phase models were compared and evaluated by its performance in classification.

• Keywords: krajolik dobrote; algoritmi optimizacije; evolucijski algoritmi; strojno učenje; neuronska mreža; genetsko programiranje; diferencijska evolucija; algoritam selekcije klona; simbolička regresija; stablo odluke
• Keywords (english): Fitness landscape; optimization algorithms; evolution algorithms; machine learning; neural network; genetic programming; differential evolution; clonal selection algorithm; symbolic regression; decision tree
• Language: croatian
• URN:NBN: urn:nbn:hr:168:898367
• 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: 2018-02-05 17:50:39