Multi-layer training algorithm of a pulse neural network

Abstract

A multi-layer training algorithm of a spiking neural network utilizes a layer-by-layer training mode to allow intra-layer connection, combines weight training and structure training, and sharpens relevance between data, and specifically comprises the steps of 1) carrying out data preprocessing: converting input data into a pulse sequence through a conversion function; 2) initializing a network layer: setting the number of layers of the pulse neural network, the number of neurons of each layer and the distribution state of neurons in the layer; 3) carrying out interlayer structure pre-training:in the initial network, no connection is set between network layers, and the interlayer connection is generated in a layer-by-layer recursion mode; 4) normalizing the interlayer weight: eliminating the influence caused by data difference through the normalization operation on the interlayer weight; 5) performing intra-layer structure training: performing intra-layer structure training by using aneural network structure training algorithm, and 6) performing network weight causality training The network trained by the algorithm has self-organizing and self-growing capabilities, and is simple in algorithm rule, small in calculation amount, low in requirement on model precision and easy to simulate.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, in particular to a multi-layer training algorithm of a pulse neural network. Background technique [0002] Artificial neural network originated from the simulation of biological neural network, bionicity is not only one of the important characteristics of artificial neural network, but also the driving force of its intelligence. The spiking neural network is currently the most biologically interpretable artificial neural network. Compared with mainstream neural networks such as forward propagation networks and deep learning networks, it has stronger bionic properties. Therefore, the study of spiking neural networks is of great importance to brain-like intelligence. significance. [0003] Data in a spiking neural network is represented in a spike-encoded manner, which makes it incompatible with the backpropagation algorithm. Effective training algorithm is the most basic and impo...

AI Technical Summary

Problems solved by technology

[0005] The technical problem solved by the present invention is: to overcome the problem that current spiking neural network multi-layer training is difficult to converge, and to set the training algorithm from the two aspects of weight and structure, thus providing a multi-layer neural network with self-organization, Self-growth training algorithm

Images