Modifications of the Interactive Learning for Competitive Associative Nets and the Application of Piecewise Linear approximation ability to Nonlinear Problems
| Project/Area Number |
16300070
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Sensitivity informatics/Soft computing
|
|---|
| Research Institution | Kyushu Institute of Technology |
|---|
Principal Investigator |
KUROGI Shuichi Kyushu Institute of Technology, Faculty of Engineering, Associate Professor, 工学部, 助教授 (40178124)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
NISHIDA Takeshi Kyushu Institute of Technology, Faculty of Engineering, Research Associate, 工学部, 助手 (30346861)
|
|---|
| Project Period (FY) |
2004 – 2006
|
| Project Status |
Completed (Fiscal Year 2006)
|
| Budget Amount *help |
¥3,600,000 (Direct Cost: ¥3,600,000)
Fiscal Year 2006: ¥1,000,000 (Direct Cost: ¥1,000,000)
Fiscal Year 2005: ¥1,000,000 (Direct Cost: ¥1,000,000)
Fiscal Year 2004: ¥1,600,000 (Direct Cost: ¥1,600,000)
|
|---|
| Keywords | Competitive Associative Nets / Modification of Interactive Learning / Piecewise Linear Approximation / Application to Nonlinear Problems / Neural Networks / Control of Nonlinear Plants / Analysis of Speech Time Series |
|---|
| Research Abstract |
The aim of this research was to modify the conventional interactive learning method for the competitive associative net (CAN2), and investigate the application of the piecewise linear approximation of the CAN2 to nonlinear problems. This research have achieved the following results.
(1) Modification and examination of the interactive learning method: At first, we have modified the interactive (online) learning method into a batch learning method because the latter scheme is effective in the case when available data are finite which often occurs in practical problems. Next, in order to minimize the generalization error stably, we have modified to ensemble learning methods which aggregate CAN2s with different number of units, and bagging learning methods which aggregate CAN2s which learn bootstrap resample datasets. The third modification was to use first-order difference of a time series for time series prediction. Through several challenges to international prediction competitions, we have achieved the third place at the time series prediction competition held at the IJCNN2004, the regression winner at Evaluating Predictive Uncertainty Challenge held at NIPS2004, and the second place at the time series prediction competition held at ESTSP2007.
(2) Application to the control of nonlinear time-varying plants: We have applied the above modified methods to the temperature control of RCA solutions which is for cleaning silicon wafers. We have achieved the improvement of the performance of the control, especially in the stability in learning process, as well as reducing the settling time and the overshoot.
(3) Application to the analysis and recognition of the speech time series: The above modified methods are applied to reproduction, recognition, and analysis of vowel signals, and we have obtained the improvement of the performance.
|
Report
(4 results)
Research Products
(40 results)
