METHOD OF PROCESSING DATA BY MEANS OF A NEURAL NETWORK SUBJECTED TO DECOMPOSITION TAKING INTO ACCOUNT THE AMOUNT OF MEMORY OF A COMPUTING DEVICE (VERSIONS), AND A COMPUTER-READABLE MEDIUM Russian patent published in 2024 - IPC G06N3/02 

FIELD: data processing.

SUBSTANCE: invention relates to methods of processing data using a neural network subjected to decomposition taking into account the available memory of a computing device. Method includes at least the following steps, at which: receiving from a peripheral device a first fragment of input data and storing it in memory of a computing device; processing said first fragment of input data by means of a first unit of a decomposed neural network, and storing the processed data in a memory area of the computing device in which the first input data fragment is stored, replacing said first input data fragment; receiving from the peripheral device a second fragment of input data and storing it in a free area of memory of the computing device; and processing said second fragment of input data by means of second unit of neural network subjected to decomposition, and storing the processed data in a memory area in which a second input data fragment is stored, replacing said second input data fragment.

EFFECT: reduced memory consumption of a computing device during execution of a neural network.

15 cl, 8 dwg