Influence of self-connection weights on cellular-neural network stability
Cellular-Neural Associative Memory (memory by Hophield with local connection structure) with weight matrix designed by anyone of the existing methods ensuring individual stability of network is concidered. It is studied how self-connection weight values influence the main characteristic of CNAM, namely the strong stability to k-distortions of stored prototypes. Expression for determining the self-connection weight values is obtained, such that provides a maximal strong stability for each prototype. Two strategies are proposed to determine the most acceptable value according to the requiered accuracy. The obtained results are valid not only for CNAM but also for full-connected Hopfield associative memory designed with the help of any learning method.
Unable to display preview. Download preview PDF.
- 1.J.J. Hopfield, D.W.Tank. Computing with Neural Circuits: a Model. Science, Vol.233, 1986, p. 625.Google Scholar
- 2.L.Perzonas, I.Guyon, G.Dreyfus. Collective Computational Properties of Neural Networks: New Learning Mechanism. Physical Review, A, vol 34, November, 1986,p. 4217–4228.Google Scholar
- 3.M.Cottrell. Stability and Attractivity in Associative Memory Networks. Biological Cybernetics 58, 1988, p. 129–139.Google Scholar
- 4.J.Zhang, Li Zhang, D.Yan, A.He, L.Liu. Local Interconnection Neural Network and its Optical Implementation. Optics Communication, Vo1.102, 1993, pp.3–20.Google Scholar
- 5.S.G.Pudov. Learning of Cellular-Neural Associative Memory. Avtometria, (to be published in N 3, 1997).Google Scholar
- 6.D.O.Gorodnichy. Desaturating Coefficient for Projection Learning Rule. Lecture Notes in Computer Sciense, 1112, 1996, p.469–476.Google Scholar
- 7.O.L.Bandman. Cellular-Neural Computations, Formal Model and Possible Applications. Lecture Notes in Computer Science, 964, 1995, p.21–35.Google Scholar