Self-organizing collaborative neural network model learning and construction method

Abstract

The invention discloses a self-organizing collaborative neural network model learning and construction method. The learning method comprises the steps of preprocessing data; initializing a network; and carrying out prototype mode self-learning based on the SOM network. The construction method comprises the following steps: calculating an adjoint mode; and after a adjoint mode matrix is calculated,using the prototype mode matrix as a weight for connecting the input layer and the sequence parameter layer, using the adjoint mode matrix as a weight for connecting the sequence parameter layer andthe output layer, and enabling the network to work. The method has the advantages that a SNN network model is improved, and the problems that the SNN structure is single and the expansibility is poorare solved through redefinition of ordered parameters of the prototype mode and the adjoint mode of the prototype mode. On the basis, a Kohonen network layer is introduced and combined with an SNN sequence parameter layer, and the problem that global regulation and control are difficult in the SNN working process is solved on the basis of the practical significance of the similarity degree of a sequence parameter prototype mode and an input sample.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence and pattern recognition, in particular to a self-organized synergistic neural network (Self-organized SNN, SoSNN) model learning and construction method based on synergetic neural network (Synergetic Neural Network, SNN). Background technique [0002] In 1973, the German physicist Haken first proposed the Synergetics theory, which has since been widely used in various fields such as physics, chemistry, biology, neuroscience, psychology and sociology. As an important branch theory of systems science, synergistics mainly studies how open systems far from equilibrium can spontaneously emerge orderly structures in time, space and function through internal synergy. Similarly, the coordination of various regions in the brain is also the main mode for the brain to solve complex problems, which is fundamentally different from the traditional ANN. The traditional ANN embodies a hier...

AI Technical Summary

Problems solved by technology

[0004] (1) The problem of single structure and poor scalability of SNN

[0005] The current collaborative neural network structure is fixed and relatively simple. The introduction of the same type of prototype mode will increase the recognition error rate, and the support for new types is poor, and it is difficult to further improve the performance of the network.

[0006] (2) SNN lacks the problem of prototype mode learning method

[0007] SNN does not have an effective prototype mode learning method, and the prototype mode is not representative of this type of sample, resulting in a low recognition rate and unable to take advantage of SNN

Images