Efficient learning system and method based on spiking neural network

Abstract

The invention discloses an efficient learning system based on a spiking neural network and a learning method thereof, and belongs to the technical field of artificial neural networks. The learning system is composed of a data preprocessing module, a network construction module, a learning module and an inference module. The data preprocessing module supports two types of static image data and dynamic neuromorphic data. The network construction module supports multiple types of spiking neural networks. The learning module is composed of feedforward network calculation, error back propagation and weight updating. At each moment, the random number generator generates a random value, and whether error back propagation and weight updating are carried out or not is controlled by a switch. The inference module loads the model file to carry out feedforward network calculation. The learning system is compatible with an automatic differential tool and is easy to implement, and various pulse neural network types are supported, and flexible configuration is easy. A switch structure is added to the learning module, a large number of calculation steps are ignored, calculation resources are saved, and the learning efficiency is improved.

Description

technical field [0001] The invention relates to the technical field of artificial neural networks, in particular to an efficient learning system and a learning method based on an impulse neural network. Background technique [0002] In recent years, the research and application of Spiking Neural Network (SNN) has become more and more extensive. Different from Artificial Neural Network (ANN), spiking neural network uses neurons similar to biological plasticity as computing units, which can better represent dynamic temporal and spatial information. The information transfer method of spiking neural network is discrete pulses, which enables it to be applied to low-power, high-efficiency neuromorphic hardware. Due to the lack of efficient learning systems, spiking neural networks still do not perform as well as artificial neural networks in many applications. [0003] The current learning algorithms of spiking neural networks can be classified into three categories: unsupervise...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the defect that the spiking neural network learning algorithm in the prior art requires relatively high computing resources and storage resources, and the learning efficiency is low, the present invention provides a learning system and a learning method based on a spiking neural network. The learning system consists of It is composed of data preprocessing module, network building module, learning module and reasoning module; the data preprocessing module supports two types of static image data and dynamic neuromorphic data; the network building module su

Images