Thesis MHTMetLib : a new library to measure the performance of trajectory metaheuristics
No Thumbnail Available
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Informática
Departament
Campus
Campus Santiago San Joaquín
Abstract
Los algoritmos metaheurísticos proponen una buena forma de solucionar problemas, al comparar con tecnicas con enfoques determinísticos o técnicas completas de búsqueda, permitiendo disminuir los tiempos de ejecución a cambio de obtener soluciones aproximadas al optimo. El problema es que su forma estocástica conlleva incerteza para medir su rendimiento, dificultando el análisis de los experimentos y el diseño de ellos. Es por ello que se propone en este trabajo diseñar e implementar una librería para medir el rendimiento de algoritmos metaheurísticos trayectoriales en problemas combinatoriales, permitiendo un análisis completo de ellos en etapas de experimentación. Se valida esta solución con un conjunto de algoritmos trayectoriales clasicos utilizando el framework IOHprofiler, permitiendo obtener un analisis completo acerca de distintas caracteristicas propias de cada algoritmo en la ejecución de distintos problemas benchmark de optimización binaria.
Metaheuristic algorithms offer a good way to solve problems when compared to deterministic approaches or complete search techniques, allowing for reduced execution times in exchange for obtaining solutions that are approximate to the optimal. The problem is that their stochastic nature brings uncertainty when measuring their performance, making it difficult to analyze experiments and design them. Therefore, this work proposes to design and implement a library to measure the performance of trajectory-based metaheuristic algorithms in combinatorial problems, allowing a comprehensive analysis of them during experimentation stages. This solution is validated with a set of classic trajectory-based algorithms using the IOHprofiler framework, enabling a complete analysis of various characteristics specific to each algorithm when solving different binary optimization benchmark problems.
Metaheuristic algorithms offer a good way to solve problems when compared to deterministic approaches or complete search techniques, allowing for reduced execution times in exchange for obtaining solutions that are approximate to the optimal. The problem is that their stochastic nature brings uncertainty when measuring their performance, making it difficult to analyze experiments and design them. Therefore, this work proposes to design and implement a library to measure the performance of trajectory-based metaheuristic algorithms in combinatorial problems, allowing a comprehensive analysis of them during experimentation stages. This solution is validated with a set of classic trajectory-based algorithms using the IOHprofiler framework, enabling a complete analysis of various characteristics specific to each algorithm when solving different binary optimization benchmark problems.
Description
Keywords
Algoritmos metaheurísticos trayectoriales, Métricas de rendimiento, Problemas de optimización, Trajectory-based metaheuristic algorithms, Performance metrics
