ANALOGUE HARDWARE IMPLEMENTATION OF NEURAL NETWORKS Russian patent published in 2023 - IPC G06N3/63 G06G7/60 

FIELD: neural networks.

SUBSTANCE: invention can be used for analogue hardware implementation of neural networks. The technical result consists in providing analogue hardware implementation of neural networks with reduced energy consumption and network load. The method includes the steps of obtaining the topology of the neural network and the weight factors of the trained neural network, transforming the topology of the neural network into an equivalent analogue network of analogue components, calculating the weight matrix for the equivalent analogue network based on the weight factors of the trained neural network, wherein each element of the weight matrix represents the corresponding connection between the analogue components of the equivalent analogue network, forming a schematic model for the implementation of the equivalent analogue network based on the weight matrix, including the choice of component values for the analogue components.

EFFECT: providing analogue hardware implementation of neural networks with reduced energy consumption and network load.

90 cl, 158 dwg