REFLECTIVE NEURAL NETWORK

摘要

FIELD: computer engineering, in particular, image recognition. SUBSTANCE: reflective network has neural nodes with multiple-level bottom-up receptor and self-reflection circuits and top-down sanctioning and effecting circuits. Said circuits represent clone hierarchical structure of vertical profiles with bottom-up and top-down axon inter-level connections. This results in possibility of correction of synaptic weights using system self-organization motivation signals together with phase detection signals which are distributed in time. This is modeled both by adding increments during learning process and correction of synaptic weights during image recognition. EFFECT: increased precision. 14 dwg