データコラボレーション解析による分散協調特徴量選択手法の研究
Project/Area Number |
22K12144
|
Research Category |
Grant-in-Aid for Scientific Research (C)
|
Allocation Type | Multi-year Fund |
Section | 一般 |
Review Section |
Basic Section 61030:Intelligent informatics-related
|
Research Institution | University of Tsukuba |
Principal Investigator |
叶 秀彩 筑波大学, システム情報系, 助教 (60814001)
|
Project Period (FY) |
2022-04-01 – 2026-03-31
|
Project Status |
Granted (Fiscal Year 2022)
|
Budget Amount *help |
¥4,160,000 (Direct Cost: ¥3,200,000、Indirect Cost: ¥960,000)
Fiscal Year 2025: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
Fiscal Year 2024: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Fiscal Year 2023: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
Fiscal Year 2022: ¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000)
|
Keywords | feature selection / active feature selection / 特徴選択 / 中間表現 / データコラボレーション |
Outline of Research at the Start |
近年のデータ取得の簡易化に伴い、データは大規模・分散化している。分散管理されているデータは情報秘匿などの観点から共有が困難であり、またデータ数の不足や偏りによるリスク因子などの重要な特徴量の学習は難しくなる。重要な特徴量を学習するために,本研究では分散データの直接的な共有を行うことなく、中間表現によるデータ統合を行うことで、分散協調特徴量選択アルゴリズムを開発する。具体的には、各機関が独自に元データの抽象化を行い、抽象化されたデータ(中間表現)を同一の潜在空間に射影し、データ統合を行うことで特徴量選択のモデルを構築する。実データによる実証実験を行い、開発する特徴量選択手法の有効性を示す。
|
Outline of Annual Research Achievements |
In this year, we mainly address the active feature selection-based method and application. Most existing feature selection methods focus on statically selecting the same informative features for each subtype and fail to consider the heterogeneity of samples which causes pattern differences in each subtype. We consider active feature selection to dynamically acquire different features in each subtype by combining the subtype classifier with the reinforcement learning (RL) agent in a cost-sensitive manner. We apply active feature selection for gene signature identification in renal cell carcinoma, which can select different gene signatures for different renal cell carcinoma subtypes.
|
Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
To explore different informative feature subsets for each subtype, we propose a novel active feature selection method and apply it to gene signature identification in renal cell carcinoma. By combining the subtype classifier with the reinforcement learning (RL) agent, our method can sequentially select the active features in each sample in a cost-sensitive manner. The application for gene signature identification in renal cell carcinoma show that our method can select different gene signatures for different renal cell carcinoma subtypes.
|
Strategy for Future Research Activity |
Next step, we will further consider distributed datasets. We will propose distributed collaborative feature learning methods and consider privacy preserving. Intermediate representation based methods and federated learning will be applied to design the distributed feature selection framework for data collaboration. We also consider to apply our method to distributed multi-view datasets for multi-view data collaboration.
|
Report
(1 results)
Research Products
(8 results)