Una herramienta software para redes neuronales recurrentes
Archivos
Fecha
2005-06-26
Autores
Salazar Martínez, Sain
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
En el presente trabajo se mencionan algunos de los modelos unificadores existentes
para redes neuronales, y se analizan las ideas esenciales para una posible plataforma
de software; se propone una arquitectura para dicha plataforma basada en procesos
concurrentes, y se desarrolla una estructura de clases que sirve de bloque constructivo
para el desarrollo futuro de la plataforma. Se diseña una herramienta que incorpora
algunas de las ideas manejadas en el enfoque unificador y se particulariza en la
implementación del algoritmo Backpropagation estándar para redes feedforward y el
algoritmo Backpropagation Through Time para redes recurrentes. Se prueba la
herramienta con bases internacionales para modelos de clasificación feedforward y con
un problema de análisis de secuencia para algoritmos recurrentes.
In the present work we mention some of existent unifier models for neuronal networks; the essentials ideas are analyzed for a possible software platform. We propose an architecture for this platform based on concurrent processes that interact one with another; additionally a structure of classes is developed that serves as constructive block for the future development of the platform. A tool is designed that it incorporates some of the ideas managed in the unifier approach and it is particularized in the implementation of the standard backpropagation algorithm for feedfoward networks and backpropagation through time algorithms for recurrent networks. The tool is proven with international bases for classification with feedforward models and with a problem of sequence analysis for recurrent algorithms
In the present work we mention some of existent unifier models for neuronal networks; the essentials ideas are analyzed for a possible software platform. We propose an architecture for this platform based on concurrent processes that interact one with another; additionally a structure of classes is developed that serves as constructive block for the future development of the platform. A tool is designed that it incorporates some of the ideas managed in the unifier approach and it is particularized in the implementation of the standard backpropagation algorithm for feedfoward networks and backpropagation through time algorithms for recurrent networks. The tool is proven with international bases for classification with feedforward models and with a problem of sequence analysis for recurrent algorithms
Descripción
Palabras clave
Modelos Unificadores, Redes Neuronales Recurrentes, Herramienta Software, Diseño, Inteligencia Artificial