Title:
INFORMATION PROCESSING APPARATUS, INFORMATION PROCESSING METHOD, AND PROGRAM
Document Type and Number:
WIPO Patent Application WO/2018/155232
Kind Code:
A1
Abstract:
The present invention pertains to an information processing apparatus, an information processing method, and a program that enable accurate quantization. According to the present invention, quantization is carried out by assuming that the distribution of values calculated through an arithmetic operation of mechanical learning is based on a prescribed probability distribution. The arithmetic operation is an arithmetic operation in deep learning, and quantization is carried out by assuming that the distribution of gradients calculated through the arithmetic operation of deep learning is based on a prescribed probability distribution. In distributed learning in which mechanical learning is carried out by a plurality of devices in a distributed manner, quantization is carried out when a value obtained through learning at one device is supplied to another device. The present invention can be applied to an apparatus that carries out mechanical learning such as deep learning in a distributed manner.
Inventors:
YOSHIYAMA KAZUKI (JP)
Application Number:
PCT/JP2018/004566
Publication Date:
August 30, 2018
Filing Date:
February 09, 2018
Export Citation:
Assignee:
SONY CORP (JP)
International Classes:
G06N99/00; G06F17/30; G06N3/08
Foreign References:
| JP2016029568A | 2016-03-03 | |||
| JPH06176000A | 1994-06-24 | |||
| JP2005234706A | 2005-09-02 | |||
| JPH05108595A | 1993-04-30 |
Other References:
JUBO YOTARO: "Speech Recognition with Deep neural Net: after its first Introduction", JOURNAL OF THE JAP SO FOR ARTIFICIAL INTELLIGENCE, vol. 31, no. 2, 1 March 2016 (2016-03-01), pages 180 - 188, XP009515665, ISSN: 2188-2266
TAKEDA RYU ET AL: "Deep Neural Network Acoustic Model Training based on Weight Boundary Model for Discrete Deep Neural Networks", 9 November 2016 (2016-11-09), pages 2 - 11, XP055614599, Retrieved from the Internet
KAMYIA R., ET AL.: "PRMU2016-122 Acceleration of discrimination calculation by Binarized-DCNN and model compression", IEICE TECHNICAL REPORT PRMU, vol. 116, no. 366, 8 December 2016 (2016-12-08), pages 47 - 52, XP009515556, ISSN: 0913-5685
See also references of EP 3588394A4
TAKEDA RYU ET AL: "Deep Neural Network Acoustic Model Training based on Weight Boundary Model for Discrete Deep Neural Networks", 9 November 2016 (2016-11-09), pages 2 - 11, XP055614599, Retrieved from the Internet
KAMYIA R., ET AL.: "PRMU2016-122 Acceleration of discrimination calculation by Binarized-DCNN and model compression", IEICE TECHNICAL REPORT PRMU, vol. 116, no. 366, 8 December 2016 (2016-12-08), pages 47 - 52, XP009515556, ISSN: 0913-5685
See also references of EP 3588394A4
Attorney, Agent or Firm:
NISHIKAWA Takashi et al. (JP)
Download PDF:
Previous Patent: SOUND SOURCE DETECTING METHOD AND DETECTING DEVICE
Next Patent: IMAGE PROCESSING DEVICE, IMAGE PROCESSING METHOD, AND IMAGE SYSTEM
Next Patent: IMAGE PROCESSING DEVICE, IMAGE PROCESSING METHOD, AND IMAGE SYSTEM