Fully automatic and scalable Bayesian model selection method for tensor decomposition

2016 Fiscal Year Final Research Report

• Project/Area Number: 15K16055
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Intelligent informatics
• Research Institution: National Institute of Advanced Industrial Science and Technology (2016); National Institute of Informatics (2015)
• Principal Investigator: Hayashi Kohei 国立研究開発法人産業技術総合研究所, 人工知能研究センター, 研究員 (30705059)
• Project Period (FY): 2015-04-01 – 2017-03-31
• Keywords: テンソル分解 / モデル選択 / ベイズ学習 / アルゴリズム

Outline of Final Research Achievements

While data analysis with tensor decomposition is demanding in various application fields, in order to obtain accurate results, it is necessary to set a parameter called rank correctly, which has been adjusted by domain experts. In this study, we solve this problem by developing an algorithm that is simple, fast and highly reliable. The method has the following appealing points: (1) domain knowledge is unnecessary, (2) the algorithm is highly scalable, and (3) fully-automatic rank selection in which the performance is theoretically guaranteed is possible. Our result may yield further expansion of the application of tensor decomposition and new scientific discovery.

Free Research Field

機械学習