Title:
観測行動選択機能を有する分類システム及び該システムを使用する分類方法
Document Type and Number:
Japanese Patent JP5847425
Kind Code:
B2
Abstract:
PROBLEM TO BE SOLVED: To provide a classification system having an observing action selecting function that automatically formulates an action strategy for quickly and accurately classifying objects.SOLUTION: A classification system, while repeating actions to observe objects, classifies the objects on the basis of observation data obtained by the observing actions. The classification system (100) comprises a classification probability determining unit (103) that determines a classification probability, which is the probability of each object to be assigned to a given class; and an action selecting unit (107) that selects the next observing action on the basis of reinforcement learning which uses a status space configured of observation data and the pertinent classification probability.
Inventors:
金 天海
菅野 重樹
山崎 翔太
菅野 重樹
山崎 翔太
Application Number:
JP2011092450A
Publication Date:
January 20, 2016
Filing Date:
April 01, 2011
Export Citation:
Assignee:
本田技研工業株式会社
学校法人早稲田大学
学校法人早稲田大学
International Classes:
G06N3/00; G06N99/00
Other References:
柴田克成,他2名,Actor-Qアーキテクチャに基づく能動認識学習システム,電子情報通信学会論文誌(情報・システムII-パターン処理),日本,社団法人電子情報通信学会,2001年 9月 1日,第J84-D-II巻, 第9号,pp.2121-2130
Attorney, Agent or Firm:
特許業務法人OFH特許事務所