A VIRTUAL LAB TO VISUALIZE THE PERFORMANCE OF THE HOPFIELD’S NEURAL NETWORK FOR ASSOCIATIVE CONTENT-ADDRESSABLE MEMORY

• Autor: Montenegro Joo, Javier
• Formato: artículo
• Fecha de Publicación: 2006
• Institución: Universidad Nacional Mayor de San Marcos
• Repositorio: Revistas - Universidad Nacional Mayor de San Marcos
• Lenguaje: español
• OAI Identifier: oai:ojs.csi.unmsm:article/8614
• Enlace del recurso: https://revistasinvestigacion.unmsm.edu.pe/index.php/fisica/article/view/8614
• Nivel de acceso: acceso abierto
• Materia: Cybernetics; Artificial Intelligence; Artificial Neural Networks; Hopfield; Pattern Recognition; Image Reconstruction; Ising; Magnetism.; Cibernética; Inteligencia Artificial; Redes de Neuronas artificiales; Reconocimiento de patrones; Reconstrucción de imágenes; Magnetismo.

A software to appreciate the performance of the Hopfield’s Neural Network (Associative Content-Addressable Memory) has been developed and applied to computer synthesized images. A rather small network has been created and applied to four sets of training-remembrances. The software allows the user to become familiar with associative memory computer simulations, it also provides knowledge on the work of a neural net, hence this computer program may be used as a training tool (teach - learn) on neural networks. This software makes evident that a straightforward application of neural networks is in the field of pattern recognition and image reconstruction; it also serves as an introduction to more advanced and complex neural nets. This report is aimed at understanding the performance and potentials of a neural network, it may also foster the interest of students in cybernetics. The Hopfield neural network is important to physicists because it is closely related to the Ising Spin Glass model of magnetism, the learned memories in the net stand for the low energy states in the Ising model. A set of images (shown in this report) is included in the software, however it also accepts those made by the user.