A log-linear modeling of unstable variable selection: improving stability and extracting multiple sets of variables

• Project/Area Number: 15K16064
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Soft computing
• Research Institution: Kobe University
• Principal Investigator: Kitazono Jun 神戸大学, 工学研究科, 工学研究科研究員 (00733677)
• Project Period (FY): 2015-04-01 – 2019-03-31
• Project Status: Completed (Fiscal Year 2018)
• Budget Amount: ¥4,030,000 (Direct Cost: ¥3,100,000、Indirect Cost: ¥930,000); Fiscal Year 2017: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000); Fiscal Year 2016: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000); Fiscal Year 2015: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
• Keywords: 変数選択 / 安定性 / 対数線形モデル / 可視化 / クラスタリング / シナジー / Minimum Probability Flow / 特徴選択 / 機械学習

Outline of Final Research Achievements

Variable selection is a process of selecting useful variables for pattern recognition tasks, such as classification and regression. Recent studies showed that variable selection is often unstable: even a small change in data results in huge difference of selected variables. This study aimed at resolving this instability of variable selection. We developed a framework for identifying truly useful variables by finding regularity among varied combinations of selected variables.

Academic Significance and Societal Importance of the Research Achievements

変数選択は，医学・工学・理学等多くの場面で使用されている．例えば，遺伝子解析では，変数選択によって，特定の病気に将来なりやすいかどうか識別するのに有用な遺伝子を選び出すことが可能である．変数選択の安定性の問題を解消し，信頼性を高めることにより，より確度の高い病気リスク診断などへの発展が期待される．また，本研究の過程で開発した可視化やクラスタリングの手法は，変数選択に限らず，多くのデータ分析に適用可能なものもあり，幅広い分野での応用が期待される．

Research Products

• [Int'l Joint Research] Universite Paris 13/Universite Paris Descartes(フランス)
• [Int'l Joint Research] Universite Paris 13/Universite Paris Descartes(フランス)
• [Journal Article] t-Distributed stochastic neighbor embedding spectral clustering (2017)
  • Author(s): N. Rogovschi, J. Kitazono, N. Grozavu, T. Omori and S. Ozawa
  • Journal Title: Proceedings of 2017 International Joint Conference on Neural Networks Volume: - Pages: 1628-1632
  • DOI: 10.1109/ijcnn.2017.7966046
  • Peer Reviewed / Int'l Joint Research
• [Journal Article] Multidimensional Unfolding Based on Stochastic Neighbor Relationship (2017)
  • Author(s): N. Murata, J. Kitazono, S. Ozawa
  • Journal Title: Proceedings of the 9th International Conference on Machine Learning and Computing Volume: 印刷中
  • Peer Reviewed
• [Journal Article] t-Distributed Stochastic Neighbor Embedding with Inhomogeneous Degrees of Freedom (2016)
  • Author(s): J. Kitazono, N. Grozavu, N. Rogovschi, T. Omori, S. Ozawa
  • Journal Title: Doya K., Ikeda K., Lee M., Liu D. (eds) Neural Information Processing. ICONIP 2016. Lecture Notes in Computer Science Volume: 9949 Pages: 119-128
  • DOI: 10.1007/978-3-319-46675-0_14
  • ISBN: 9783319466743, 9783319466750
  • Peer Reviewed / Int'l Joint Research / Acknowledgement Compliant
• [Journal Article] An Exhaustive Search and Stability of Sparse Estimation for Feature Selection Problem (2015)
  • Author(s): K. Nagata, J. Kitazono, S. Nakajima, S. Eifuku, R. Tamura, M. Okada
  • Journal Title: IPSJ Online Transactions Volume: 8 Issue: 0 Pages: 25-32
  • DOI: 10.2197/ipsjtrans.8.25
  • NAID: 130005091237
  • ISSN: 1882-6660
  • Peer Reviewed
• [Presentation] Multidimensional Unfolding Based on Stochastic Neighbor Relationship (2017)
  • Author(s): N. Murata, J. Kitazono, S. Ozawa
  • Organizer: The 9th International Conference on Machine Learning and Computing
  • Place of Presentation: Nanyang Executive Centre, Singapore
  • Year and Date: 2017-02-24
  • Int'l Joint Research
• [Presentation] t-Distributed Stochastic Neighbor Embedding Spectral Clustering (2017)
  • Author(s): N. Rogovschi, J. Kitazono, N. Grozavu, T. Omori and S. Ozawa
  • Organizer: 2017 International Joint Conference on Neural Networks
  • Int'l Joint Research
• [Presentation] t-Distributed Stochastic Neighbor Embedding with Inhomogeneous Degrees of Freedom (2016)
  • Author(s): J. Kitazono, N. Grozavu, N. Rogovschi, T. Omori, S. Ozawa
  • Organizer: The 23rd International Conference on Neural Information Processing
  • Place of Presentation: 京都大学（京都府・京都市）
  • Year and Date: 2016-10-16
  • Int'l Joint Research
• [Presentation] Machine Learning for Making Data Understandable (2016)
  • Author(s): Jun Kitazono
  • Organizer: Cyber-Physical System for Smarter World 2016
  • Place of Presentation: Seaside Hotel Maiko Villa Kobe
  • Year and Date: 2016-03-23
  • Int'l Joint Research / Invited
• [Presentation] Minimum Probability Flowによるt分布型確率的近傍埋め込み法の高速化 (2015)
  • Author(s): 北園淳，大森敏明，小澤誠一
  • Organizer: 第59会システム制御情報学会研究発表講演会
  • Place of Presentation: 中央電気倶楽部（大阪）
  • Year and Date: 2015-05-20