US 12,142,030 B2
Neural network compression device and method for same
Daichi Murata, Tokyo (JP)
Assigned to HITACHI ASTEMO, LTD., Ibaraki (JP)
Appl. No. 17/781,510
Filed by Hitachi Astemo, Ltd., Hitachinaka (JP)
PCT Filed Oct. 30, 2020, PCT No. PCT/JP2020/040924
§ 371(c)(1), (2) Date Jun. 1, 2022,
PCT Pub. No. WO2021/111788, PCT Pub. Date Jun. 10, 2021.
Claims priority of application No. 2019-220172 (JP), filed on Dec. 5, 2019.
Prior Publication US 2023/0005244 A1, Jan. 5, 2023
Int. Cl. G06V 10/774 (2022.01); G05D 1/00 (2024.01); G06N 3/082 (2023.01); G06V 10/82 (2022.01)
CPC G06V 10/774 (2022.01) [G05D 1/0088 (2013.01); G06N 3/082 (2013.01); G06V 10/82 (2022.01)] 15 Claims
OG exemplary drawing
 
1. A neural network compression device that compresses a neural network by using a training data set that has been input, the neural network compression device comprising:
a training image selection unit that calculates an influence value on an inference result calculated by using an inference data set and a neural network model for the training data set, that classifies the training data set into valid training data necessary for the compression and invalid training data unnecessary for the compression, based on the influence value, and that generates an indexed training data set; and
a neural network compression unit that compresses the neural network model, based on the indexed training data set and the neural network model.