| Project/Area Number |
08640639
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Physical chemistry
|
|---|
| Research Institution | Ochanomizu University |
|---|
Principal Investigator |
NAGASHIMA Umpei Ochanomizu University, Faculty of Science, Professor, 理学部, 教授 (90164417)
|
|---|
| Project Period (FY) |
1996 – 1997
|
| Project Status |
Completed (Fiscal Year 1997)
|
| Budget Amount *help |
¥1,800,000 (Direct Cost: ¥1,800,000)
Fiscal Year 1997: ¥500,000 (Direct Cost: ¥500,000)
Fiscal Year 1996: ¥1,300,000 (Direct Cost: ¥1,300,000)
|
|---|
| Keywords | neural network / structure-acitivity correlation / self-organized network model / ^<13>C-NMR shift / 再構築 |
|---|
| Research Abstract |
In 1996, a perceptron type neural network simulator for structure-activity corpelation of molecules has been developped with two different pre-education methods.
As an ecample of application, conformations of norbornene isomers were predicted using ^<13>C-NMR data. The predicted conformations are in excellent agreement with experimetns And it was suggested that ^<13>C-NMR data of only two specified carbons in the norbornene skeleton have strong correlation with the conformation of the main branch.
In 1997, a self-organized network model for high-speed learning was included in the neural network simulator. The performance of the self-organized network model was compared with that of perceptron using two- and three-dimensional exclusive OR problem ant the relationship between ^<13>C-NMR shift and the conformation of norbornane. In the case of ^<13>C-NMR shift and conformation of norbornane, a self-organized network seems to show strong nonlinear classification in comparsion with perceptron. 38
An artificial neural network simulation was also applied to the recognition and reproduction of time series data whose amplitudes and frequenciessimultaneously change with time. The results of our model were compared with those obtained by the least squares method using four kinds of model functions. Our model gives higher quality results than the least squares method especially in the prediction of amplitude change.
|
