2005 Fiscal Year Final Research Report Summary
Hardware Realization of Neural Oscillator with Learning Capability
| Project/Area Number |
16500142
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Sensitivity informatics/Soft computing
|
|---|
| Research Institution | Kansai University |
|---|
Principal Investigator |
MAEDA Yutaka Kansai University, Faculty of Engineering, Professor, 工学部, 教授 (60209393)
|
|---|
| Project Period (FY) |
2004 – 2005
|
| Keywords | neural oscillator / FPGA / learning / simultaneous perturbation / hardware implementation / FPAA |
|---|
| Research Abstract |
In this research, I propose a learning scheme for pulse coupled oscillators using the simultaneous perturbation optimization method and its hardware implementation. It was difficult and complicated for usual optimization method to find proper parameter values of the pulse coupled oscillator, since the oscillator is a kind of recurrent neural network. The simultaneous perturbation method gives a simple solution. Moreover, this approach is suitable for hardware realization. From this point of view, I proposed and fabricated the hardware pulse coupled oscillator and recurrent neural network with learning ability via the simultaneous perturbation method.
First of all, I confirm feasibility of the proposed pulse coupled oscillator with learning capability through simulation by C language and MatLab. The pulse coupled oscillator can generate pulse train with desired interval through leaning process.
Hardware realization of neural networks is an interesting issue. Mainly there are two approache
… More
s ; digital realization and analog one. As the former approach, field programmable gate array(FPGA) is useful target. I designed the pulse coupled oscillator with learning capability by VHDL. Then, design result is configured on FPGA. I verified the operation of the FPGA pulse coupled oscillator system with learning ability. Proper pulse train is obtained.
The second approach is analog realization. Then, field programmable analog array(FPAA) is adopted to implement the oscillator. Analog circuit design of the system is carried out. The circuit operation is confirmed by a circuit simulator. The system could realize the pulse coupled oscillator with learning capability via the simultaneous perturbation method. Next, the design is configured to FPAA. The circuit realized the pulse coupled oscillator. Interval of generated pulse train varied depending on parameters contained in the pulse coupled oscillator.
Moreover, some recurrent neural networks with learning capability were realized by FPGA using the simultaneous perturbation method. Hopfield network and bidirectional associative memory are typical examples of the recurrent networks. Usually, it was difficult to realize these hardware recurrent neural network systems with learning capability. However, I implemented the Hopfield neural network system and bidirectional neural network system with learning ability using the simultaneous perturbation method. I showed some application of these systems.
As a result, I could confirm a validity and feasibility of the pulse coupled oscillator with learning capability via the simultaneous perturbation method. These systems were fabricated and tested the operation of these systems. Less
|
