| Project/Area Number | 13680453 |
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Intelligent informatics
|
|---|
| Research Institution | Kyushu Institute Of Technology |
|---|
Principal Investigator |
SHO Hiroshi Kyushu Institute Of Technology, Department of brain Science & Engineering, Associate Professor, 大学院・生命体工学研究科, 助手 (30235709)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
ISHIKAWA Masumi Kyushu Institute Of Technology, Department of brain Science & Engineering, Professor, 大学院・生命体工学研究科, 教授 (60222973)
|
|---|
| Project Period (FY) |
2001 – 2003
|
| Project Status |
Completed(Fiscal Year 2003)
|
| Budget Amount *help |
¥3,500,000 (Direct Cost : ¥3,500,000)
Fiscal Year 2003 : ¥600,000 (Direct Cost : ¥600,000)
Fiscal Year 2002 : ¥700,000 (Direct Cost : ¥700,000)
Fiscal Year 2001 : ¥2,200,000 (Direct Cost : ¥2,200,000)
|
|---|
| Keywords | inverse optimization / two-point reaching movement / Discrete-time quadratic constraints / interpretation of time-series data / 離散時間2次形制約条件 / ニューラルネットワーク / 学習 / 二次形制約 / 時系列データ解析 |
|---|
| Research Abstract |
In this research, we proposed various structures of neural network which represent optimality conditions regarding inverse optimization problems. Due to find out a criterion function of decision-maker underlying given time series data of two-point reaching movement, we established some methods of interpretation data by real world data. In our investigations, a numerical value model that reflects spatiotemporal features of given time series data can be obtained by a method for generating discrete time quadratic constraints. Since the noise of the movement command is included in the give data, a scatter about the data occurrence at time step. For preventing it, it is necessary to obtain an average length of the given data by complementing data each other. Accordingly, the obtained numerical value model could be used for solve inverse optimization problems as a known constraints. Thereupon, based on the method of discrete-time static inverse optimization, we can estimate a criterion function under which given time series data become optimal subject to the constraints. By this way, it is confirmed to estimate a criterion function that makes the given time series data of two-point reaching movement become optimal by learning of neural networks. This, approach indicates that it is possible to interpret a difference between each model for analyzing the kinematics characteristic of human arm. It is left for study to compare results between the solution of dynamic inverse optimization problem and the solution of the static inverse optimization problem.
|
