| Project/Area Number |
12834002
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Institution | Hirosaki University |
|---|
Principal Investigator |
MIYANO Takaya Department of Intelligent Machines and System Engineering, Hirosaki University, Associate Professor, 理工学部, 助教授 (10312480)
|
|---|
| Project Period (FY) |
2000 – 2001
|
| Project Status |
Completed (Fiscal Year 2001)
|
| Budget Amount *help |
¥2,100,000 (Direct Cost: ¥2,100,000)
Fiscal Year 2001: ¥700,000 (Direct Cost: ¥700,000)
Fiscal Year 2000: ¥1,400,000 (Direct Cost: ¥1,400,000)
|
|---|
| Keywords | nonlinear systems / complex systems / local approximation / time series analysis / classification / support vector algorithm |
|---|
| Research Abstract |
1. Nonlinear local approximation method for time series prediction, regression and classification of discrete sequences of data is developed based on the Sugihara-May method. The approximation model is capable of learning the metric to measure the distance between input vectors with the Caprile-Girosi algorithm.
2. A data base of medicine and welfare is constructed with aid of the Ministry of Health, Welfare and Labor for numerical experiment of examining applicability to nursing care needs certification.
3. A data base of speech signals is constructed with aid of the Aomori Radio and Broadcast Corporation.
4. A data base of atomic motion in solid crystals as numerical solutions of molecular dynamics simulation is constructed with aid of Sumitomo Metal Industries, Ltd.
5. From numerical experiments of predicting total care time and total nursing time with the medical data base, it is shown that the nonlinearity of local hyper surface representing the functional dependence between situation
… More
s of patients and the total nursing care time can be measured in terms of the optimized metric. This implies that the complexity of the decision rule or the underlying dynamics of complex behavior can be estimated by the metric proposed. The experiment also exhibits that coarse-graining of objective variables may improve the predictive performance of the approximate, although the nonlinearity of the local hyper surface is enhanced by the coarse-graining.
6. The approximating method is successfully applied to speech signals and numerical atomic behavior to reproduce the dynamics underlying the time series data.
7. The proposed metric could be utilized as an index associated with the Vapnik-Chervonenkis dimension to control the generalization error of local approximates. Clarifying the mathematical relationship between the metric and the VC (dimension is remained to be investigated in the future. The present work may be able to be said to providing a key ingredient for building a variation of the support vector algorithm for local nonlinear approximation techniques. Less
|
