Neuron calculation method

Abstract

The invention relates to the field of neuron calculation, in particular to a neuron calculation method. The method comprises the following steps: calculating a plurality of input neuron membrane potentials Vin (t1) and input neuron pulses Pin (t1) by neurons within a time interval t, updating the state of the neurons, and sending output neuron membrane potentials Vout (t2) and output neuron pulses Pout (t2); the unit time of the time interval t is delta t; the starting moment of the time interval is recorded as t1 = n1 * delta t, and the ending moment of the time interval is recorded as t2 = n2 * delta t; the time interval t is equal to n * delta t, and n is equal to n2-n1; comprising the following steps: acquiring input data at t1 moment; within the time interval t, performing integration or suppression operation on the input data; within the time interval t, executing membrane potential accumulation operation; within the time interval t, executing membrane potential activation operation; and at the t2 moment, updating the neuron state and outputting data. The neuron calculation method provided by the invention has better universality, so that a more universal neural network calculation system is realized.

Description

technical field [0001] The invention relates to the field of neuron computing, in particular to a neuron computing method that can be used in numerical neural networks and impulse neural networks. Background technique [0002] Neuron cells are the basic functional units for building the biological nervous system of the human brain. Numerous neuron cells are connected to each other in a complex way to build an efficient intelligent perception and cognition system. A typical biological neuron cell consists of three parts, namely synapse / dendrite, cell body and axon. [0003] The working process of biological neuron cells is roughly as follows: the synapse / dendritic is the input device of the neuron cell, which is connected with many other neuron cells, and is responsible for collecting and transmitting signals from other neuron cells; the cell body is the neuron cell's A computing device that performs computations on all input signals; an axon is the output device of a neuron...

AI Technical Summary

Problems solved by technology

Therefore, existing neuron models cannot directly process spiking neural networks converted from numerical neural networks, nor can they directly process spiking data from dynamic sensors.

[0007] Second, in the existing method, the neuron can only realize the integration operation of the input signal

The integration of input signals refers to the cumulative calculation of all input signals, and in the numerical neural network algorithm, there is also an input signal suppression operation (taking the maximum value of the input signal), so the existing neuron model cannot directly handle the transformation by the numerical neural network. The resulting spiking neural network

[0008] Third, in existing methods, neurons can only output pulse signals

The pulse signal only exists in the pulse neural network, and the data of the numerical neural network is usually an 8-bit integer or 16 / 32-bit floating point number, so the existing neuron model cannot directly handle the numerical neural network algorithm

[0009] Existing neuron models only support spiking neural networks, and cannot be directly used for numerical neural network calculations, dynamic sensors, or spiking neural networks converted from numerical neural networks.

Images