Primjena evolucijskih algoritama u latentnom prostoru modela dubokog učenja

master's thesis

• Title (english): Evolutionary algorithm optimization in latent space of deep learning models
• Author: Maria Katić
• Mentor: Domagoj Jakobović (mentor)
• Committee member: Domagoj Jakobović (predsjednik povjerenstva)
• Committee member: Nikolina Frid (član povjerenstva)
• Committee member: Leonardo Jelenković (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-09, Croatia

Abstract

U ovom diplomskom radu istražena je primjena evolucijskih algoritama u latentnom prostoru modela varijacijskog autoenkodera (VAE). Za demonstraciju rada algoritma korišten je optimizacijski problem trgovačkog putnika (TSP). Implementacija rada temelji se na korištenju varijacijskog autoenkodera koji uči iz primjeraka i rješenja TSP problema te kreira latentni prostor koji se zatim koristi za efikasno pretraživanje rješenja. U sklopu rada napravljena je detaljna analiza hiperparametara koji utječu na performanse algoritma te usporedba različitih optimizacijskih algoritama unutar latentnog prostora. Analiza je pokazala da korištenje kombinacije genetskog algoritma i metode lokalne pretrage Hooke-Jeeves u latentnom prostoru postiže značajne rezultate prilikom pronalaska optimalnog rješenja za dani problem. Rezultati istraživanja ukazuju na prednosti i mane optimizacije u latentnom prostoru i potiču na daljnje istraživanje tog područja.

Abstract (english)

This thesis explores the application of evolutionary algorithms in the latent space of the variational autoencoder (VAE) model, specifically looking at solving the Traveling Salesman Problem (TSP). The implementation of the work is based on the use of a variational autoencoder that learns from examples and solutions of TSP problems and creates a latent space that is then used for an efficient search for solutions. As part of the work, a detailed analysis of the hyperparameters that influence the performance of the algorithm and a comparison of different optimization algorithms within the latent space was made. The analysis showed that the use of a combination of the genetic algorithm and the Hooke-Jeeves local search method in the latent space achieves significant results when finding the optimal solution for a given problem. The research results indicate the advantages and disadvantages of optimization in the latent space and encourage further research in this area.

• Keywords: duboko učenje; varijacijski autoenkoder; latentni prostor; genetski algoritam; diferencijska evolucija; Hooke-Jeeves postupak; problem trgovačkog putnika
• Keywords (english): deep learning; variational autoencoder; latent space; genetic algorithm; differential evolution; Hooke-Jeeves method; traveling salesman problem
• Language: croatian
• URN:NBN: urn:nbn:hr:168:964349
• 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: 2024-10-12 21:02:14