Study on the data-driven fusion of nonparametric and sparse modeling

2018 Fiscal Year Final Research Report

• Project/Area Number: 16K16108
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Intelligent informatics
• Research Institution: The Institute of Statistical Mathematics (2018); University of Tsukuba (2016-2017)
• Principal Investigator: Hino Hideitsu 統計数理研究所, モデリング研究系, 准教授 (10580079)
• Project Period (FY): 2016-04-01 – 2019-03-31
• Keywords: 機械学習 / ノンパラメトリックモデル / スパースモデリング

Outline of Final Research Achievements

Throughout the research period, we developed several component technologies to realize the concept of the fusion of nonparametric and sparse modeling. We developed a non-parametric mixture distribution estimation method in the framework of information geometry, an intrinsic dimension estimation method based on the generalized linear modeling. We have developed an image super-resolution method and an estimation method of latent graph structure. We also applied conventional and tailored methods to several fields of science. Particularly we collaborated with geoscientists as originally planned to analyse slow slip earthquakes and classification of magma tectonic fields. We also collaborated with material scientists and developed a method for accelarating the X-ray spectroscopy microscope measurement. Also, through the cooperation with researchers in electrophysiological measurements, we proposed a method for estimation of nerve cell connections based on partially observed signals.

Free Research Field

機械学習

Academic Significance and Societal Importance of the Research Achievements

古典的なデータ解析・統計手法が開発された時点とは質・量ともに異なるデータを扱うための新たなデータ解析手法の開発は，工学にとっても科学にとっても喫緊の課題である．データに特定の分布を仮定しないノンパラメトリックモデルは計算・メモリコストが高い．大規模高次元データに適用可能な疎表現アプローチはモデルとして柔軟さに欠ける．本研究課題では，ノンパラメトリックモデルとスパースモデルの融合を目指し，ノンパラメトリック手法，スパースモデリング手法それぞれで新たな手法を開発し，地球科学，材料科学，情報科学の諸問題に適用して学術的知見の発見，計測技測度の向上を実現した．