2016 Fiscal Year Final Research Report
Manifold signal processing theory concerning metric structure and its application to biological signal processing
| Project/Area Number |
26280054
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Partial Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Research Field |
Perceptual information processing
|
|---|
| Research Institution | Tokyo Institute of Technology |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
杉山 将 東京大学, 新領域創成科学研究科, 教授 (90334515)
田中 聡久 東京農工大学, 工学(系)研究科(研究院), 准教授 (70360584)
鷲沢 嘉一 電気通信大学, 情報理工学(系)研究科, 助教 (10419880)
|
|---|
| Project Period (FY) |
2014-04-01 – 2017-03-31
|
| Keywords | 多様体信号処理 / 局所独立方程式 / 局所等方独立方程式 / 計量学習 / 生体信号処理 / ブレイン・コンピュータ・インターフェース |
|---|
| Outline of Final Research Achievements |
For the manifold signal processing theory concerning the metric structure, we constructed a method to solving locally isotropically independence equations by using Gauss normalized kernel functions, and defined an operator equation to express the locally isotropically independent change. We developed a semi-supervised learning method for metrics when the number of training data is not enough. We proposed a method to assign a pattern to a subspace, and a metric suitable to discrimination for the Grassmann manifold. We also developed an anomaly detection method with images by combining the direct estimation for the ration between probability density functions and the convolutional network, and a distance for hand-written character recognition that provides the best recognition rate among those based on k-nearest neighbor method. Furthermore, we improved the performance of recognition of motional intention that enables paretics caused by apoplexy, etc. to reconstruct motor function.
|
| Free Research Field |
パターン認識・画像処理
|
