Method and system for realizing competitive learning mechanism of spiking neural network based on memristor

Abstract

The invention discloses a method and system for realizing a competitive learning mechanism of a spiking neural network based on a memristor, and the method comprises the steps: enabling an activated neuron of a learning layer in the spiking neural network to receive a pulse from an input layer, and carrying out the learning under an STDP rule; the method further comprises a step of performing lateral suppression through a synapse realized based on a memristor and a suppression-type neuron connected with one or more activation-type neurons: if the suppression-type neuron is activated by currentfrom the synapse, enabling the suppression-type neuron to send a suppression-type pulse to the connected activation-type neuron through the synapse so as to realize lateral suppression. According tothe invention, the memristor is used as a synapse to realize a transverse suppression and balance mechanism, the expandability of the network can be improved, and the inherent device advantages of thememristor are effectively utilized to reduce the hardware complexity and power consumption of the unsupervised SNN network.

Description

technical field [0001] The invention relates to a hardware implementation technology of a pulse neural network, in particular to a method and system for realizing a competitive learning mechanism of a pulse neural network based on a memristor. Background technique [0002] Spiking neural network (SNN) is a bionic neural network model, which has broad prospects in realizing low-power, high-efficiency intelligent systems. [0003] The basic constituent units of SNN include spiking neuron and synapse. Spiking neuron is a computing unit, and synapse is a connection between spiking neurons and a channel for transmitting information between neurons. The carrier of the neuron is the spike signal, which is sent by one neuron and transmitted to another neuron through the synapse. The pulse neuron receives pulse signals from other pulse neurons through the synapse, and obtains a new neuron state through the calculation of its own dynamic equation. When the state variable of the neuro...

AI Technical Summary

Problems solved by technology

[0012] However, existing methods for achieving lateral inhibition require connections proportional to the square of the number of learning neurons, limiting the scalability of the network; existing hardware solutions for balancing mechanisms require each learning neuron to have complex circuitry to regulate neuron threshold, which creates costly hardware complexity

Images