Neuron information sending method and system

Abstract

The invention relates to a neuron information sending method. The method includes: determining a neuron demultiplexing group, wherein the neuron demultiplexing group comprises at least two neurons; configuring a demultiplexing rule of a current neuron according to the neuron demultiplexing group, wherein the demultiplexing rule divides an operation step into at least two operation periods, the quantity of the operation periods is greater than or equal to the quantity of the neurons, and the neurons in the neuron demultiplexing group respectively correspond to the operation periods; and outputting neuron information in the corresponding operation period thereof by the neurons in the neuron demultiplexing group in each operation period of the current operation step according to the demultiplexing rule. According to the method, the neuron information is sent in an ordered manner according to the set operation period so that the neurons can send more information in the duration of the current operation step, the capability of information sending of the neurons is improved, and the information processing capability of the whole neural network is improved.

Description

technical field [0001] The invention relates to the technical field of neural networks, in particular to a neuron information sending method and system. Background technique [0002] Neuromorphic engineering was proposed by Carver Mead in 1990, which aims to use large-scale integrated circuits to simulate the architecture of biological nervous systems and build neuromorphic computing systems. Early neuromorphic computing systems were generally implemented by analog circuits, but in recent years digital circuits and digital-analog hybrid circuits have been increasingly used by neuromorphic engineering. At present, neuromorphic engineering and neuromorphic circuits are one of the emerging research hotspots in the world. Traditional neuromorphic computing platforms aim to simulate brain neuron models and ion channel activities through analog circuits, and use digital circuits and on-chip storage to build connections and routes, so that it is very convenient to change the neuro...

AI Technical Summary

Problems solved by technology

When carrying neural network operations, this limits the output of each layer of the neural network to no more than 256, that is, the number of neurons in the next layer cannot exceed 256, that is, in the traditional neural network, the connection between neurons is limited, Greatly limit the information processing ability of the neural network

Images