A Study of A Classification Method Using Automatic Model Selection For Multimedia Heterogeneous Data
| Project/Area Number |
16300036
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Media informatics/Database
|
|---|
| Research Institution | The Institute of Statistical Mathematics |
|---|
Principal Investigator |
MATSUI Tomoko The Institute of Statistical Mathematics, Research Organization and Systems, Associate Professor (10370090)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
TANABE Kunio Waseda University, Faculty of Science and Engineering, Professor (50000203)
|
|---|
| Project Period (FY) |
2004 – 2007
|
| Project Status |
Completed (Fiscal Year 2007)
|
| Budget Amount *help |
¥12,360,000 (Direct Cost: ¥11,400,000、Indirect Cost: ¥960,000)
Fiscal Year 2007: ¥4,160,000 (Direct Cost: ¥3,200,000、Indirect Cost: ¥960,000)
Fiscal Year 2006: ¥2,100,000 (Direct Cost: ¥2,100,000)
Fiscal Year 2005: ¥2,800,000 (Direct Cost: ¥2,800,000)
Fiscal Year 2004: ¥3,300,000 (Direct Cost: ¥3,300,000)
|
|---|
| Keywords | multimedia / classification / model selection / probabilistic model / inductive learning / カーネルマシン / 音声認識 / 話者認識 / 画像認識 |
|---|
| Research Abstract |
Our objective is to develop a practical classification method bawd on dual penalized regression machines (dPLRMs) for heterogeneous data such as multimedia data including speech, image and text data. dPLRMs are multiclass discrimination machines which has been developed by a investigator of this project, Kunio Tanabe and can handle noisy stochastic data by employing the penalized logistic regression model. We have worked on four specific problems; 1) establishment of a practical classification framework using dPLRMs, 2) investigation on modeling of heterogeneous, variable-length, and massive data sets, 3) investigation of a method to deal with unknown data, and 4) development of the dPLRM software package for wide use. For 1), we established the framework through the experiments to examine the classification and inductive power in dPLRMs with speech and auditory data. For 2), we extended dPLRMs for multiple kernels and investigated a coding method to manage heterogeneous data with different sampling rates. Moreover, we designed a kernel function for time series with variable lengths. For 3), we investigated a method to add a new class for unknown data. For 4), we developed the software package and opened it to the public with a research purpose. Through the 1)-4) investigations, we established the fundamental framework of a classification method based on dPLRMs.
|
Report
(5 results)
Research Products
(57 results)
