FAULT-TOLERANT NEURAL NETWORK, METHOD FOR DETECTING NEURAL NETWORK FAULTS, METHOD FOR ENSURING FAULT TOLERANCE OF NEURAL NETWORK Russian patent published in 2021 - IPC G06F11/20 G06N3/02 

FIELD: computer technology.

SUBSTANCE: fault-tolerant neural network is disclosed, consisting of two types of elements, of neurons that perform functions of information processing and decision-making, and switches for transmitting information, wherein a switch consists of a table of mutual connections of inputs and outputs and a device that transmits information between its inputs and outputs based on this table, where inputs and outputs of the switch are connected to inputs and outputs of other switches and inputs and outputs of neurons of the network, wherein neurons of the network have one input and one output, inputs and outputs of a neuron are connected to inputs and outputs of the switch, while at least one redundant neuron is connected to each switch, switches are connected to each other by inputs and outputs, and they form a multiconnected network, in which there is at least one redundant path between each pair of switches, and there is at least one redundant switch, and there is also at least one redundant upper-level switch that is connected to inputs and outputs of the neural network and inputs and outputs of other switches.

EFFECT: providing a fault-tolerant neural network.

5 cl, 1 dwg