US 12,333,429 B2
Partial inference framework for sequential DNN processing on constrained devices, and acoustic scene classification using said partial inference framework
Sven Ewan Shepstone, Struer (DK); and Pablo Martínez Nuevo, Fredriksberg (DK)
Assigned to Bang & Olufsen A/S, Struer (DK)
Filed by BANG & OLUFSEN A/S, Struer (DK)
Filed on Oct. 29, 2021, as Appl. No. 17/514,107.
Prior Publication US 2022/0138571 A1, May 5, 2022
Int. Cl. G06N 3/08 (2023.01); G06N 3/04 (2023.01); G10L 25/30 (2013.01); G10L 25/51 (2013.01)
CPC G06N 3/08 (2013.01) [G06N 3/04 (2013.01); G10L 25/30 (2013.01); G10L 25/51 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method for performing inference on input data using a neural network, wherein said neural network comprises at least two neural network layers, a first neural network layer and a subsequent second neural network layer, the method comprising the steps of:
obtaining input data;
storing the input data in a data storage arrangement;
obtaining parameter data indicating the parameters of the first neural network layer;
storing the parameter data of the first layer in a parameter data storage location of the data storage arrangement;
processing said input data using the stored first layer parameter data, to form first layer output data;
storing said first layer output data in the data storage arrangement;
obtaining parameter data indicating the parameters of the second neural network layer and storing the second layer parameter data by replacing the first layer parameter data with the second layer parameter data in the parameter data storage location;
processing said first layer output data using the stored second layer parameter data to form second layer output data; and
storing said second layer output data in the data storage arrangement,
wherein the parameter data of said at least two neural network layers is obtained by receiving the parameter data from an external device over a light-weight messaging scheme.