A framework for inverse modeling applied to multi-objective evolutionary algorithms

Oliveira, Artur Leandro da Costa

Por favor, use este identificador para citar o enlazar este ítem: https://ri.ufs.br/jspui/handle/riufs/16597

Registro completo de metadatos

Campo DC	Valor	Lengua/Idioma
dc.contributor.author	Oliveira, Artur Leandro da Costa	-
dc.date.accessioned	2022-10-10T13:49:10Z	-
dc.date.available	2022-10-10T13:49:10Z	-
dc.date.issued	2022-06-08	-
dc.identifier.citation	OLIVEIRA, Artur Leandro da Costa. A framework for inverse modeling applied to multi-objective evolutionary algorithms. 2022. 144 f. Dissertação (Mestrado em Ciência da Computação) – Universidade Federal de Sergipe, São Cristóvão, 2022.	pt_BR
dc.identifier.uri	http://ri.ufs.br/jspui/handle/riufs/16597	-
dc.language	eng	pt_BR
dc.subject	Computação	por
dc.subject	Aprendizado do computador	por
dc.subject	Linguagem unificada de modelagem	por
dc.subject	Unified modeling language (UML)	eng
dc.subject	Multi-objective optimization	eng
dc.subject	Machine learning	eng
dc.subject	Inverse models	eng
dc.title	A framework for inverse modeling applied to multi-objective evolutionary algorithms	pt_BR
dc.type	Dissertação	pt_BR
dc.contributor.advisor1	Carvalho, André Britto de	-
dc.description.resumo	Many-Objective Optimization Problems (MaOPs) are a class of complex optimization problems deined by having more than three objective functions. Traditional Multi-Objective Evolutionary Algorithms (MOEAs) have shown poor scalability in solving this kind of problem. The use of machine learning techniques to enhance optimization algorithms applied to MaOPs has been drawing attention due to their capacity to add domain knowledge during the search process. One method of this kind is inverse modeling, which uses machine learning models to enhance MOEAs diferently, mapping the objective function values to the decision variables. This method has shown a good performance in diverse optimization problems due to the ability to directly predict solutions closed to the Pareto-optimal front, among these methods, we can highlight the Decision Variable Learning (DVL). The strategies involving inverse models found, including the DVL, have some limitations such as the exploration of the performance of diferent machine learning models and the strategies in using the generated knowledge during the search. The main goal of this work is to create a framework that uses an inverse modeling approach coupled to any MOEA found in the literature. More precisely, three main steps were taken to achieve the goals. First, we perform a systematic review of the literature to identify the main uses of machine learning techniques enhancing optimization algorithms. Secondly, we analyze the performance of diferent machine learning methods in the DVL, seeking to understand the main characteristics of inverse modeling through the DVL algorithm. In the last step, we propose a framework that is an extension of the DVL algorithm, based on the knowledge obtained in the systematic review and our analysis of the DVL. This framework results in an algorithm for MaOPs recommended for situations that exist restrictions on the number of evaluations in the objective function.	pt_BR
dc.publisher.program	Pós-Graduação em Ciência da Computação	pt_BR
dc.subject.cnpq	CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO	pt_BR
dc.publisher.initials	Universidade Federal de Sergipe	pt_BR
dc.contributor.advisor-co1	Gusmão, Renê Pereira de	-
dc.description.local	São Cristóvão	pt_BR
Aparece en las colecciones:	Mestrado em Ciência da Computação

Ficheros en este ítem:

Fichero	Descripción	Tamaño	Formato
ARTUR_LEANDRO_COSTA_OLIVEIRA.pdf		3,71 MB	Adobe PDF	Visualizar/Abrir

Mostrar el registro sencillo del ítem