| Project/Area Number |
12650411
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
System engineering
|
|---|
| Research Institution | University of the Ryukyus |
|---|
Principal Investigator |
KINJO Hitoshi University of the Ryukyus, Department of Engineering, Associate Professor, 工学部, 助教授 (50211206)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
NAKAZONO Kunihiro University of the Ryukyus, Department of Engineering, Research Assistant, 工学部, 助手 (80284959)
YAMAMOTO Tetsuo University of the Ryukyus, Department of Engineering, Professor, 工学部, 教授 (20045008)
|
|---|
| Project Period (FY) |
2000 – 2001
|
| Project Status |
Completed (Fiscal Year 2001)
|
| Budget Amount *help |
¥800,000 (Direct Cost: ¥800,000)
Fiscal Year 2001: ¥400,000 (Direct Cost: ¥400,000)
Fiscal Year 2000: ¥400,000 (Direct Cost: ¥400,000)
|
|---|
| Keywords | Neural networks / Emotional processing / Autonomous system / Moving robots / Behavior control / Spiking neural networks / Genetic algorithms |
|---|
| Research Abstract |
The main objectives of this research project are to construct emotional neural networks and apply them in robot control.
In this project, we develop spiking neural networks (SNNs) using genetic algorithms (GA). In a previous study, we developed emotional neural networks using a model of an analog neuron, by computer simulation. However, neural networks of living organisms are based on electrical spikes. It is said that information processing in the brain occurs through neuron activation by neural spikes. Therefore a model of an emotional neural network should also be based on neuron activation by spikes.
In engineering fields, many training methods for artificial neural networks have been reported. However, there are few reports concerning SNNs. In this study, we present a training method for SNNs using GA. The back-propagation (BP) method, which is a very popular and powerful training method, cannot be easily applied to SNN training because spiking of a neuron is a discontinuous neural activity function. The GA method can be used to successfully train the SNN independently of the mode of the activity function. In the simulation study, we confirmed that the GA method is suitable for SNN training and that SNNs can be easily applied as emotional neural networks.
The application of emotional neural networks based on the SNN model to robot control is left for future work.
|
