Title:
CONVOLUTIONAL NEURAL NETWORK
Document Type and Number:
WIPO Patent Application WO/2019/008951
Kind Code:
A1
Abstract:
Provided is a convolutional neural network (20) including a first convolutional layer (22, 24), a pooling layer (28, 29), and a second convolutional layer (23, 25), and comprising: a crossbar circuit (44) provided with a plurality of input bars (50), output bars (51, 52), and a plurality of weighting elements (53); and a control unit (40). In the crossbar circuit, each input value inputted via the input bars is assigned weight while also being added in the output bars and outputted from the output bars, said weighting corresponding to a position within a second filter of a pooling computation result computed from each of the input values. The cross bar circuit is also provided with a division part (54a) and is configured such that an average pooling computation vis-à-vis a convolutional computation result of the first convolutional layer and convolutional computation of the second filter vis-à-vis the average pooling computation result are carried out simultaneously.
Inventors:
KATAEVA IRINA (JP)
OTSUKA SHIGEKI (JP)
OTSUKA SHIGEKI (JP)
Application Number:
PCT/JP2018/020712
Publication Date:
January 10, 2019
Filing Date:
May 30, 2018
Export Citation:
Assignee:
DENSO CORP (JP)
International Classes:
G06N3/063; G06G7/60; G06T7/00
Foreign References:
| US20160284400A1 | 2016-09-29 | |||
| JP2017078934A | 2017-04-27 |
Other References:
YAKOPCIC, CHRIS ET AL.: "Memristor Crossbar Deep Network Implementation Based on a Convolutional Neural Network", 2016 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN, 11 June 2018 (2018-06-11), pages 963 - 970, XP032992268, Retrieved from the Internet
Attorney, Agent or Firm:
JIN Shunji (JP)
Download PDF: