Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
多層ニューラルネットの大局構造抽出装置、方法、及びプログラム
Document Type and Number:
Japanese Patent JP6770480
Kind Code:
B2
Abstract:
To compress a multilayer neutral net.SOLUTION: A whole situation structure extraction part extracts the community of a layer d for each layer d of a multilayer neural net based on a coupling relationship between peaks of adjacent layers which are determined by each side in a multilayer neutral net after learning. A side definition part defines a side that connects between communities of the adjacent layers based on the community of each layer d extracted by the whole situation structure extraction part, and when the community having no side to the next layer exists, the peak included in the community and the side connected to the peak included in the community are deleted from a neural net, and when no side exists between two communities cand c, the side existing between the communities cand cis deleted from the neural net.SELECTED DRAWING: Figure 8

Inventors:
Chihiro Watanabe
Application Number:
JP2017100256A
Publication Date:
October 14, 2020
Filing Date:
May 19, 2017
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
Nippon Telegraph and Telephone Corporation
International Classes:
G06N3/08
Domestic Patent References:
JP200242107A
JP5108595A
Attorney, Agent or Firm:
Patent Service Corporation Taiyo International Patent Office