Algoritmo para optimizar la topología en un Mapa Cognitivo Difuso
Fecha
2014-07-06
Autores
Pérez García, Ricardo
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
Los Mapas Cognitivos Difusos (MCD) son una técnica de computación blanda que constituye una prometedora metodología de modelado, especialmente útil para modelar y simular sistemas dinámicos. De forma general éstos se pueden definir como grafos pesados dirigidos con retroalimentación, consistentes en nodos y de relaciones causales entre los nodos. Éstos fueron introducidos por Kosko y desde entonces han surgido numerosos algoritmos de aprendizaje supervisados y no supervisados, los cuales han estado principalmente enfocados en la estimación de la causalidad del sistema (matriz de pesos causales) usando datos históricos y/o la intervención de expertos, mostrando buenos resultados. Estos algoritmos de aprendizaje pueden ser catalogados en tres grupos: algoritmos hebbianos, algoritmos poblacionales y algoritmos híbridos. A pesar de su éxito, estos poseen ciertas limitantes. Es válido mencionar que hasta donde el autor tiene conocimiento no existen algoritmos de aprendizaje para MCD que permitan encontrar el subconjunto mínimo de nodos centrales que describa el sistema modelado de una forma eficiente. Con este objetivo proponemos un algoritmo de aprendizaje discreto basado en técnicas de Inteligencia Colectiva. El comportamiento de esta propuesta es estudiado a través de un caso de estudio real concerniente a la predicción de la resistencia a fármacos de proteínas del VIH.
Fuzzy Cognitive Maps (FCM) is a soft computing technique which constitute a promising modeling methodology, especially useful for modeling and simulating dynamic systems. In general, they can be defined as directed weight graphs with feedback, consisting in nodes and causal relations among the nodes. Since its introduction, many supervised and not supervised learning algorithms have been proposed, mainly focused in estimated the system causality (causal weight matrix) using historic data and/or expert’s intervention, showing good results. These learning algorithms can be classified in three groups: hebbian-based algorithms, population-based algorithms and hybrid approaches. Despite their success, they have some limitations. It is valid to mention that as far as the author knowledge, there not exist learning algorithms for FCM that allows finding the subset of central nodes describing the modeled system in an efficient way. With this goal in mind, we propose a discrete learning algorithm based in Swarm Intelligence techniques. The proposal's behavior is studied through a real study case related to the field of drug resistance prediction in HIV.
Fuzzy Cognitive Maps (FCM) is a soft computing technique which constitute a promising modeling methodology, especially useful for modeling and simulating dynamic systems. In general, they can be defined as directed weight graphs with feedback, consisting in nodes and causal relations among the nodes. Since its introduction, many supervised and not supervised learning algorithms have been proposed, mainly focused in estimated the system causality (causal weight matrix) using historic data and/or expert’s intervention, showing good results. These learning algorithms can be classified in three groups: hebbian-based algorithms, population-based algorithms and hybrid approaches. Despite their success, they have some limitations. It is valid to mention that as far as the author knowledge, there not exist learning algorithms for FCM that allows finding the subset of central nodes describing the modeled system in an efficient way. With this goal in mind, we propose a discrete learning algorithm based in Swarm Intelligence techniques. The proposal's behavior is studied through a real study case related to the field of drug resistance prediction in HIV.
Descripción
Palabras clave
Algoritmos de Aprendizaje, Inteligencia Colectiva, Mapas Cognitivos Difusos, Optimización, Topología, Caso de Estudio, Resistencia a los Fármacos, Proteínas del VIH, Bioinformática