US 11,657,267 B2
Neural network apparatus, vehicle control system, decomposition device, and program
Mitsuru Ambai, Tokyo (JP)
Assigned to DENSO IT LABORATORY, INC., Tokyo (JP)
Appl. No. 16/318,779
Filed by Denso IT Laboratory, Inc., Tokyo (JP)
PCT Filed Jul. 20, 2017, PCT No. PCT/JP2017/026363
§ 371(c)(1), (2) Date Jun. 6, 2019,
PCT Pub. No. WO2018/016608, PCT Pub. Date Jan. 25, 2018.
Claims priority of application No. JP2016-143705 (JP), filed on Jul. 21, 2016.
Prior Publication US 2019/0286982 A1, Sep. 19, 2019
Int. Cl. G06F 7/38 (2006.01); G06N 3/08 (2023.01); G06F 17/16 (2006.01); G06N 3/04 (2023.01); G06N 3/063 (2023.01); G06N 3/045 (2023.01); G06N 3/048 (2023.01)
CPC G06N 3/08 (2013.01) [G06F 17/16 (2013.01); G06N 3/04 (2013.01); G06N 3/045 (2023.01); G06N 3/048 (2023.01); G06N 3/063 (2013.01)] 7 Claims
OG exemplary drawing
 
1. A neural network apparatus comprising:
a storage unit storing a neural network model; and
an arithmetic unit inputting input information into an input layer of the neural network model and outputting an output layer,
wherein a weight matrix of at least one layer of the neural network model is constituted by a product of an integer matrix serving as a weight basis matrix and a real number matrix serving as a weight coefficient matrix,
wherein, in the at least one layer, the arithmetic unit uses an output vector from a previous layer as an input vector to decompose the input vector into a sum of a product of an integer matrix serving as an input basis matrix and a real number vector serving as an input coefficient vector and an input bias and derives a product of the input vector and the weight matrix,
wherein the weight basis matrix is a binary matrix or a ternary matrix, and the input basis matrix is a binary matrix, and
the arithmetic unit conducts product operation between the weight basis matrix and the input basis matrix with use of logical operation and bit count.