Lightweight on-chip learning method and system based on spiking neural network, and processor

Abstract

The invention belongs to the technical field of microprocessors, and particularly discloses a lightweight on-chip learning method based on a spiking neural network. The method comprises the steps: carrying out the rate coding of an input image, converting a static frame image into a pulse form, taking each pixel point as a presynaptic neuron, and enabling an output layer of the spiking neural network to be composed of LIF neuron, wherein each neuron is a post-synaptic neuron, the pre-synaptic neuron and the post-synaptic neuron are connected in a full connection mode, and in the training of the spiking neural network, the weight of each synaptic is updated according to an STDP weight updating learning rule; when the post-synaptic neurons emit pulses, the neurons are winning neurons, the winning neurons are taken as centers, membrane potentials of other post-synaptic neurons are inhibited according to the chessboard distances of the post-synaptic neurons, and finally the self-organizing pulse neural network is obtained. The problem that the shallow pulse network recognition rate is low is solved.

Description

technical field [0001] The invention belongs to the technical field of microprocessors, and in particular relates to a light-weight on-chip learning method, system and processor based on a pulse neural network. Background technique [0002] The rapid development of traditional deep neural network has made it widely used in various fields of society, including face recognition, smart home, smart medical care, etc. Dealing with problems such as slow speed and low energy efficiency. Although it has the support of a GPU or a dedicated accelerator chip, it cannot break through the bottleneck of energy efficiency, and it is difficult to realize on-chip learning, and the cost also increases accordingly. On the other hand, the brain-like mechanism that simulates the pulse processing of the human brain, that is, the pulse neural network, encodes information through pulses and performs transmission processing. The problem of low recognition rate of shallow spiking network, while dee...

AI Technical Summary

Problems solved by technology

[0003] One of the purposes of the present invention is to provide a light-weight on-chip learning method based on spiking neural networks, which solves the problem of low recognition rate of shallow spiking networks, and inherits the asynchronous sparse calculation of spiking networks with simple operation, Advantages of fast speed, high energy efficiency, and support for on-chip learning

Images