INFORMATION ENCODING METHOD IN PULSED NEURAL NETWORKS Russian patent published in 2021 - IPC G06N3/02 

FIELD: simulation of pulsed neurons.

SUBSTANCE: invention can be used for image recognition, analysis and image processing. Method includes stages, on which as a unit information symbol for encoding whole partial pulse sequence from among all input ones or its absence is used; generating clusters of partial pulse sequences selectively tuned to position significant input information; pulse sequences, which correspond to informatively insignificant input information are assumed to be zero; instantaneous summation of instantaneous values of all partial pulse sequences of each of formed selective clusters is performed and obtaining in each recording neuron pulse sequences with alternating amplitude peaks, equal to the sum of peaks of partial pulse sequences, excitation with peaks of pulses of summary sequences a threshold waiting self-oscillator, which converts peaks of total pulse sequences at its input into pulses or pulse sequence at output; if the intensity of the object at the input varies, then the "average frequency" of all partial sequences at the input is changed synchronously, which leads to change of "average frequency" of pulse sequence at output of pulse self-oscillator, equal to almost-period of partial pulse sequences at its input.

EFFECT: providing invariance of coding location of objects and intensity of input signal, high stability of operation.

1 cl, 18 dwg