A Self Organizing MAP And Adaptive Resonance Theory Learning Synergetics Neural Network

Artificial intelligence has become one of the most rapidly developing subjects in the field of pattern recognition applications.As a self-organizing artificial neural network,the synergetics neural network has unique advantages in the interpretability and model visualization.This thesis is based on the theory of synergy and other selforganization theory,using self-organizing mapping network and adaptive resonance theory,proposes a new type of synergetics neural network model based on SOM and ART learning,which can solve open,dynamic and complex problems.The main contents of this article are as follows:(1)In order to solve the traditional SNN structure is fixed,and with poor scalability,combined with the Kohonen layer structure in the self-organizing mapping network,an improved collaborative neural network model is proposed.In this part,we study the new network structure design pattern and the definition of parameters in the network.In order to verify the validity of the model,combined with a specific example of MNIST image dataset classification,the training mode and operational procedure of the network are studied.The convergence of the network and the model classification results obtained by the network are analyzed.We also studied the intrinsic link between the hidden layer topology and the prototype model learned by the network.The results of our proposed improved model on the MNIST dataset can reach the same level of current advanced machine learning classifiers.(2)Based on the previous research,this paper proposes a new class discovery technology based on adaptive resonance theory for the lack of continuous learning ability of SNN and the discovery of new network classes.At the same time,the hierarchical growth model of the network is studied.The network structure allows the network to adaptively grow under the parameter control to the hierarchical structure required for certain specific problem.Compared with the previous model,the proposed hierarchical model has lower time complexity and improves the performance of the network.We have carried out experiments on the EMNIST handwritten character dataset.Due to the local optimization method,the proposed method obtained a better result was,the accuracy of classification reached 91.15%.(3)For the classification problem of large sample complex features,combined with convolutional neural network,a synergetics neural network model supported by CNN is proposed to classify image convolution features.Firstly,after learning the data with CNN,the pre-trained convolution layer is used to extract the convolution characteristics of the image,and the convolution features are trained and classified in the synergetics neural network.By verifying the synergetics neural network model supported by CNN on the ImageNet dataset,the proposed model improves the recognition rate on the dataset compared with other models,and verifies the validity of the model,our method raised accuracy by 5 %.Using the new synergistic neural network as a classifier,this paper researching intelligent,self-organizing and adaptive image classification models,using CNN for feature optimization and extraction,can solve open,dynamic and complex problems well and provide new ideas and solutions for technical development of artificial intelligence.