1992 Fiscal Year Final Research Report Summary
A Fundamental Research of Artificial Neural Network with the Ability of Intensive Information Processing
| Project/Area Number |
03650302
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
情報工学
|
|---|
| Research Institution | Teikyo University |
|---|
Principal Investigator |
TAGUCHI Hideo Teikyo Univ., sch. of Sci. & Engin, Assoc.Prof., 理工学部, 助教授 (40029278)
|
|---|
| Project Period (FY) |
1991 – 1992
|
| Keywords | Intensive Information Processing / Artificial Neuron / Pulse Transmission / Analog-MOS Circuit / Learning Circuit / Neuro-Board / Neural Network / Information Mapping |
|---|
| Research Abstract |
The theme of this project was to proceed a fundamental research of artificial neurons' assembly that might equal a brain in high intensive information processing. This research was discussed in both sides of hardware and software. Research results are summarized as follows.
1. In this research, it was kept in mind that intelligent activities of a living body would depend on the intensive information processing (i.e.,fusion/integration of sensory informations) in a brain. Besides, the advanced information processing by artificial neural networks were discussed. Hence, it was especially suggested that the dynamics of artificial neural network itself should be considered to perform the control information processing in a real time.
2. The target artificial neural network might be built up by the assembly of artificial neurons with each plastic activity function(output function). Thereupon, by using CAD system, a PFM (i.e., Pulse Frequency Modulation) type artificial neuron was consisted of an analogous MOS circuit. And, a neuro-board was experimentally produced in a discrete circuit. Also, a response characteristics of this neuro-board was analyzed. As the results, it was confirmed that a form of an activity function could change according to an input pulse frequency.
3. The various activity functions of neuro-board could be simulated with meromorphic curves. By standing on this simulation, a mathematical neuron with a meromorphic activity function was proposed. Furthermore, 3 layers neural network was constructed of the mathematical neurons' assembly in the EWS. By regarding this network as a tool of the information intensification, the fusion/integration of mutual informations were attempted. Consequently, it was verified that this network could be applied to various mathematical transformations and nonlinear control strategies.
|
