Construction method of neural network architecture for simulating dendritic spine change

Abstract

The invention provides a construction method of a neural network architecture for simulating dendritic spine change, which comprises the following steps: simulating a dendritic spine in the brain of a higher animal during birth by using a neural network, storing the weight of the neural network by using an adjacent matrix, and generating a weight matrix; initializing the weight matrix, simulating the pruning process of dendritic spines in the brain during growth and development of higher animals, and genrating the initialized weight matrix; obtaining training samples, and dividing the training samples into a plurality of groups, wherein each group comprises the same number of training samples; inputting each of the plurality of groups into the initialized weight matrix for training, simulating the learning process of higher animals, and generating a trained weight matrix; and converting the trained weight matrix into a real network architecture, wherein the real network architecture represents the dendritic spines of the brain of the higher animal after learning. The method can be used for a supervised image recognition task, can train a proper neural network architecture for different problems, and has high adaptability.

Description

technical field [0001] The present application relates to the technical field of computer algorithms, in particular to a method and device for constructing a neural network architecture that simulates changes in dendritic spines. Background technique [0002] 1. Current status and disadvantages of artificial neural network field: [0003] At present, with the development of deep learning, the number of neural networks with different architectures is blowing out, and most of the artificial neural networks at this stage are layered structures. For example, the convolutional neural network uses a structure such as a convolutional layer, a pooling layer, and a fully connected layer. The recurrent neural network introduces the concept of timing, but still does not get rid of the layered structure of the neural network. At present, scholars have proposed architectures such as DenseNet to connect the input layer to any subsequent layer, but the essence is still a layered structur...

AI Technical Summary

Problems solved by technology

[0007] For this reason, the first purpose of this application is to propose a method for constructing a neural network architecture that simulates changes in dendritic spines, which solves the need to artificially adjust the network architecture according to different problems in the existing artificial neural network to obtain better training results , and the reason for the adjustment is not interpretable. At the same time, it solves the problem that once the architecture of the existing artificial neural network is determined, it is difficult to adjust it during the training process. The difference between different network architectures also needs to be trained separately, and A huge time-consuming problem, by simulating the growth and regression of dendritic spines and a series of changes during the learning task to design a biologically interpretable network neural network architecture, which can be applied to supervised Image recognition tasks, and can be directly transformed in different network architectures, and a more suitable neural network architecture can also be trained for different problems, with strong adaptability

Images