Feedforward neural network structure self-organization method based on neuron saliency

Abstract

The invention discloses a self-organization method of a feedforward neural network structure based on neuron salience, relates to the technical field of neural networks, and aims at the problem of difficult selection of a traditional feedforward neural network structure, the method proposed by the invention is based on hidden layer neurons The significance size dynamically adjusts the network structure of the neural network. Experimental results show that the improved algorithm can reduce the blindness of network structure selection, realize the dynamic optimization and adjustment of network structure, and improve the accuracy of network recognition. It has high accuracy in nonlinear system identification, data classification and engineering defect type identification.

Description

technical field [0001] The invention relates to the technical field of neural networks, in particular to a feedforward neural network structure self-organization method based on neuron saliency. Background technique [0002] Artificial neural network is a research hotspot in the field of artificial intelligence since the 1980s. He abstracted the human brain neuron network from the perspective of information processing, established a mathematical model similar to the human brain, and formed different networks according to different connection methods. As a kind of error backpropagation network, feedforward neural network is the most widely used network in the field of artificial neural network research. [0003] Feedforward neural network is a kind of neural network in which information is propagated forward and error is reversed; it is a multilayer neural network with three or more layers, and each neuron is connected to each neuron on the right. Fully connected, but there...

AI Technical Summary

Problems solved by technology

[0005] 1. Local minimum problem: The traditional BP algorithm is an optimization method of local search. It needs to solve a complex nonlinear problem. The weight of the network is gradually adjusted along the direction of local improvement, which will make the algorithm Trapped in a local extremum, the weight converges to a local minimum point, which leads to network training failure

[0006] 2. Slow convergence speed: The BP algorithm is essentially a gradient descent algorithm. The objective function it needs to optimize is very complex, and a large amount of training data makes the BP algorithm inefficient, resulting in slow network convergence.

[0007] 3. The choice of neural network structure is different: there is no unified and complete theoretical guidance for the selection of neural network structure, and generally it can only be selected by experience.

[0008] 4. Poor versatility: The traditional neural network structure is unique, that is, it has a one-to-one correspondence with the training data, which reduces the versatility of the BP neural network and limits the development of the feedforward neural network.

Images