Numerical implementation of gradient algorithms

  • Authors:

  • Miguel Atencia;Yadira Hernández;Gonzalo Joya;Francisco Sandoval

  • Affiliations:

  • Departamento de Matemática Aplicada, Universidad de Málaga, Spain;Facultad de Matemática y Computación, Universidad de La Habana, Cuba;Facultad de Matemática y Computación, Universidad de La Habana, Cuba,Departamento de Tecnología Electrónica, Universidad de Málaga, Spain, Málaga, Spain;Departamento de Tecnología Electrónica, Universidad de Málaga, Spain, Málaga, Spain

  • Venue:
  • IWANN'13 Proceedings of the 12th international conference on Artificial Neural Networks: advences in computational intelligence - Volume Part II
  • Year:
  • 2013

Quantified Score

Hi-index 0.00

Visualization

Abstract

A numerical method for computational implementation of gradient dynamical systems is presented. The method is based upon the development of geometric integration numerical methods, which aim at preserving the dynamical properties of the original ordinary differential equation under discretization. In particular, the proposed method belongs to the class of discrete gradients methods, which substitute the gradient of the continuous equation with a discrete gradient, leading to a map that possesses the same Lyapunov function of the dynamical system, thus preserving the qualitative properties regardless of the step size. In this work, we apply a discrete gradient method to the implementation of Hopfield neural networks. Contrary to most geometric integration methods, the proposed algorithm can be rewritten in explicit form, which considerably improves its performance and stability. Simulation results show that the preservation of the Lyapunov function leads to an improved performance, compared to the conventional discretization.