Title:
ANALYZING DEVICE, METHOD, AND PROGRAM
Document Type and Number:
WIPO Patent Application WO/2019/176731
Kind Code:
A1
Abstract:
The present invention enables the structure of a neural network to be quantitatively analyzed.
For each combination of an input data dimension and a cluster, an analyzing portion calculates, as a relationship between the combinations, the sum of squared errors between the output of each unit belonging to a cluster when the value of the input data dimension is replaced with the average value of input data dimensions included in learning data, and the output of each unit belonging to the cluster relating to the input data prior to replacement, and for each combination of the cluster and an output data dimension, the analyzing portion calculates, as a relationship between the combinations, the squared error between the value of the output data dimension when the output value of each unit belonging to the cluster is replaced with the average output value of each unit of the cluster into which the input data included in the learning data were input, and the output data dimension value prior to replacement.
Inventors:
WATANABE CHIHIRO (JP)
HIRAMATSU KAORU (JP)
KASHINO KUNIO (JP)
HIRAMATSU KAORU (JP)
KASHINO KUNIO (JP)
Application Number:
PCT/JP2019/009114
Publication Date:
September 19, 2019
Filing Date:
March 07, 2019
Export Citation:
Assignee:
NIPPON TELEGRAPH & TELEPHONE (JP)
International Classes:
G06N3/08; G06V30/10
Foreign References:
| JPH07146852A | 1995-06-06 |
Other References:
WATANABE, CHIHIRO ET AL.: "Modular Representation of Layered Neural Networks", NEURAL NETWORKS, vol. 97, January 2018 (2018-01-01), pages 62 - 73, XP055640766, Retrieved from the Internet [retrieved on 20190418]
WATANABE, CHIHIRO ET AL.: "Global Structure Extraction Based on Positive and Negative Connection Weights in Multilayered Neural Nets", FORUM ON INFORMATION TECHNOLOGY, 5 September 2017 (2017-09-05), pages 71 - 78
SPORNS, OLAF ET AL.: "Classes of Network Connectivity and Dynamics", COMPLEXITY, vol. 7, no. 1, 15 February 2002 (2002-02-15), pages 28 - 38, XP055640769, Retrieved from the Internet [retrieved on 20190418]
WATANABE, CHIHIRO ET AL.: "Global Structure Extraction Based on Positive and Negative Connection Weights in Multilayered Neural Nets", FORUM ON INFORMATION TECHNOLOGY, 5 September 2017 (2017-09-05), pages 71 - 78
SPORNS, OLAF ET AL.: "Classes of Network Connectivity and Dynamics", COMPLEXITY, vol. 7, no. 1, 15 February 2002 (2002-02-15), pages 28 - 38, XP055640769, Retrieved from the Internet
Attorney, Agent or Firm:
TAIYO, NAKAJIMA & KATO (JP)
Download PDF: