| Project/Area Number |
10480065
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Intelligent informatics
|
|---|
| Research Institution | University of Tsukuba |
|---|
Principal Investigator |
HIRAI Yuzo UNIVERSITY OF TSUKUBA, INSTITUTE OF INFORMATION SCIENCES AND ELECTRONICS, PROFESSOR, 電子・情報工学系, 教授 (80114122)
|
|---|
| Project Period (FY) |
1998 – 1999
|
| Project Status |
Completed (Fiscal Year 1999)
|
| Budget Amount *help |
¥5,000,000 (Direct Cost: ¥5,000,000)
Fiscal Year 1999: ¥1,300,000 (Direct Cost: ¥1,300,000)
Fiscal Year 1998: ¥3,700,000 (Direct Cost: ¥3,700,000)
|
|---|
| Keywords | NEURAL NETWORKS / PRINCIPAL COMPONENT ANALYSYS / SELF-ORGANIZING LEARNING ALGORITHM / FPGA / PULSE DENSITY / DIGITAL NEURON / パルス密度型ディジタル回路 |
|---|
| Research Abstract |
A learning network which can perform PCA (Principal Component Analysis) was designed by using PDM (Pulse Density Modulating) digital circuits and fabricated in a FPGA (Field Programmable Gate Array). It was verified that the circuit could learn principal components in an input signal real-time. The discrete PCA learning algorithm proposed by Sanger was changed into a continuous form and the circuit solved a set of learning differential equations in a parallel and continuous manner.
Experiment was performed by using a small circuit with two microphone inputs and two speaker outputs. The following results were obtained :
a) After learning sound appeared from only one speaker, which extracted a principal component, when a sound source was presented in front of the microphones.
b) The sets of connection weights from the two microphones to the principal speaker converged to a principal vector which directed to the sound source.
c) When the sound source moved, the weight vector continuously tracked the sound direction in real-time.
|
