METHOD AND DEVICE FOR NEURAL TIME CODING, TRAINING AND RECOGNITION Russian patent published in 2016 - IPC G06N3/04 G06N3/08 

FIELD: data processing.

SUBSTANCE: group of inventions relates to neural networks and can be used for neural time coding, training and recognition. Method comprises steps of: utilising a relative delay line abstraction which delays one or more synaptic inputs into a neuron circuit by time delays; applying a dynamic spiking model to determine spiking behaviour of neuron circuit based on weighted and delayed synaptic inputs of neuron circuit; and adjusting, according to an unsupervised learning rule, weights associated with synaptic inputs depending on timing relation of neuron circuit spiking and delayed synaptic inputs; selecting a reusable synapse associated with neuron circuit, based on a weight of reusable synapse being reduced by unsupervised learning rule to a value below a threshold; and modifying at least one of a weight, a delay, or an afferent of reusable synapse.

EFFECT: reduced complexity of coding.

13 cl, 38 dwg