| Project/Area Number |
18H03291
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 61030:Intelligent informatics-related
|
|---|
| Research Institution | Kyushu University |
|---|
Principal Investigator |
Takeuchi Junichi 九州大学, システム情報科学研究院, 教授 (80432871)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
三村 和史 広島市立大学, 情報科学研究科, 教授 (40353297)
村田 昇 早稲田大学, 理工学術院, 教授 (60242038)
長岡 浩司 電気通信大学, 大学院情報理工学研究科, 名誉教授 (80192235)
|
|---|
| Project Period (FY) |
2018-04-01 – 2021-03-31
|
| Project Status |
Completed (Fiscal Year 2020)
|
| Budget Amount *help |
¥17,550,000 (Direct Cost: ¥13,500,000、Indirect Cost: ¥4,050,000)
Fiscal Year 2020: ¥4,290,000 (Direct Cost: ¥3,300,000、Indirect Cost: ¥990,000)
Fiscal Year 2019: ¥4,160,000 (Direct Cost: ¥3,200,000、Indirect Cost: ¥960,000)
Fiscal Year 2018: ¥9,100,000 (Direct Cost: ¥7,000,000、Indirect Cost: ¥2,100,000)
|
|---|
| Keywords | 記述長最小原理 / MDL原理 / 深層学習 / 情報理論 / MRI / サイバーセキュリティ / 教師あり学習 / 汎化誤差 / スパース学習 / 磁気共鳴画像法 / 超解像 / 汚染ガウスモデル / 進化系統樹 / 圧縮センシング / 情報幾何学 / Markov連鎖 / メタ学習 / スパース重ね合わせ符号 / Markovモデル / 誤り訂正符号 / 局所指数族バンドル / 非負値行列因子分解 / 確率的コンプレキシティ / 機械学習 |
|---|
| Outline of Final Research Achievements |
We studied development of learning theory based on the Minimum Description Length (MDL) principle and its application. Concerning the MDL principle, after reconsidering our previous results on enhancement of Barron and Cover theory (BC theory) to supervised learning, we considered the relation to deep learning. We also studied the connection between the BC theory and the stochastic complexity and its application to non-exponential families, including the mixture families and the simple contaminated Gaussian location families. As for real application, we studied MRI image reconstruction based on deep learning and data analysis for caber security. For the former topic, we proposed a high speed reconstruction method which enjoyed good image quality for MR Angiography. For the latter topic, we developed a clustering method based on the MDL principle and phylogenetic trees and showed that its performance was good by experiment using real IoT malware data.
|
| Academic Significance and Societal Importance of the Research Achievements |
様々な分野で高い実用性を示している深層学習の理論を確立することを目的に，機械学習の基盤的理論の一つであるMDL原理からのアプローチを試みている．本課題の遂行中に，MDL原理を教師あり学習に用いるための重要な条件が明らかになった．現在この観点に基づいた深層学習の解析を進めており，その結果有用な知見が得られると予想される．
|
