公开(公告)号：US11657267B2

公开(公告)日：2023-05-23

申请号：US16318779

申请日：2017-07-20

Applicant: Denso IT Laboratory, Inc.

Inventor： Mitsuru Ambai

IPC: G06F7/38 ,  G06N3/08 ,  G06F17/16 ,  G06N3/04 ,  G06N3/063 ,  G06N3/045 ,  G06N3/048

CPC classification number: G06N3/08 ,  G06F17/16 ,  G06N3/04 ,  G06N3/045 ,  G06N3/048 ,  G06N3/063

Abstract: A neural network apparatus (20) includes a storage unit (24) storing a neural network model, and an arithmetic unit (22) inputting input information into an input layer of the neural network and outputting an output layer. A weight matrix (W) of an FC layer of the neural network model is constituted by a product of a weight basis matrix (Mw) of integers and a weight coefficient matrix (Cw) of real numbers. In the FC layer, the arithmetic unit (22) uses an output vector from a previous layer as an input vector (x) to decompose the input vector (x) into a product of a binary input basis matrix (Mx) and an input coefficient vector (cx) of real numbers and an input bias (bx) and derives a product of the input vector (x) and a weight matrix (W).

公开(公告)号：US20180336430A1

公开(公告)日：2018-11-22

申请号：US15947009

申请日：2018-04-06

Applicant: Denso IT Laboratory, Inc.

Inventor： Ikuro Sato ,  Mitsuru Ambai ,  Hiroshi Doi

IPC: G06K9/46 ,  G06T7/10 ,  G06N3/04 ,  G06K9/20

CPC classification number: G06K9/46 ,  G06K9/209 ,  G06K9/6257 ,  G06N3/0454 ,  G06N3/0481 ,  G06N3/063 ,  G06N3/084 ,  G06T7/10 ,  G06T2207/20084

Abstract: A recognition system includes: a sensor processing unit (SPU) that performs sensing to output a sensor value; a task-specific unit (TSU) including an object detection part that performs an object detection task based on the sensor value and a semantic segmentation part that performs a semantic segmentation task based on the sensor value; and a generic-feature extraction part (GEU) including a generic neural network disposed between the sensor processing unit and the task-specific unit, the generic neural network being configured to receive the sensor value as an input to extract a generic feature to be input in common into the object detection part and the semantic segmentation part.

公开(公告)号：US10769479B2

公开(公告)日：2020-09-08

申请号：US15947009

申请日：2018-04-06

Applicant: Denso IT Laboratory, Inc.

Inventor： Ikuro Sato ,  Mitsuru Ambai ,  Hiroshi Doi

IPC: G06K9/00 ,  G06K9/46 ,  G06K9/62 ,  G06N3/04 ,  G06N3/063 ,  G06N3/08 ,  G06T7/10 ,  G06K9/20

Abstract: A recognition system includes: a sensor processing unit (SPU) that performs sensing to output a sensor value; a task-specific unit (TSU) including an object detection part that performs an object detection task based on the sensor value and a semantic segmentation part that performs a semantic segmentation task based on the sensor value; and a generic-feature extraction part (GEU) including a generic neural network disposed between the sensor processing unit and the task-specific unit, the generic neural network being configured to receive the sensor value as an input to extract a generic feature to be input in common into the object detection part and the semantic segmentation part.

公开(公告)号：US20190286982A1

公开(公告)日：2019-09-19

申请号：US16318779

申请日：2017-07-20

Applicant: Denso IT Laboratory, Inc.

Inventor： Mitsuru Ambai

IPC: G06N3/08 ,  G06F17/16 ,  G06N3/04

Abstract: A neural network apparatus (20) includes a storage unit (24) storing a neural network model, and an arithmetic unit (22) inputting input information into an input layer of the neural network and outputting an output layer. A weight matrix (W) of an FC layer of the neural network model is constituted by a product of a weight basis matrix (Mw) of integers and a weight coefficient matrix (Cw) of real numbers. In the FC layer, the arithmetic unit (22) uses an output vector from a previous layer as an input vector (x) to decompose the input vector (x) into a product of a binary input basis matrix (Mx) and an input coefficient vector (cx) of real numbers and an input bias (bx) and derives a product of the input vector (x) and a weight matrix (W).