Development of the learning algorithm based on feature space geometry

2016 Fiscal Year Final Research Report

• Project/Area Number: 25330276
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Intelligent informatics
• Research Institution: National Institute of Advanced Industrial Science and Technology
• Principal Investigator: Akaho Shotaro 国立研究開発法人産業技術総合研究所, 人間情報研究部門, 研究グループ長 (40356340)
• Project Period (FY): 2013-04-01 – 2017-03-31
• Keywords: 情報幾何 / カーネル法

Outline of Final Research Achievements

In order for machine learning to work effectively, it is very important to accurately capture the distance between input signals. In this research project, we tried to improve machine learning algorithm by introducing natural structure into space based on the framework of information geometry and reproducing kernel Hilbert space theory. As applications, the effectiveness was confirmed by applying to the problem of transfer learning which improves learning accuracy by synthesizing the results of similar learning problems when the number of data is small, and the problem of estimating the dimension of low dimensional structure embedded in high dimension space.

Free Research Field

機械学習