Title:
PROCESSOR FOR NEURAL NETWORK
Document Type and Number:
WIPO Patent Application WO/1993/008538
Kind Code:
A1
Abstract:
A process for neural network by which even neural networks having complex structures such as neocognitron can be made into hardwares. In a processor by which a neural network of feedforward type (e.g., neocognitron) having a multilayer structure is realized, the computation element corresponding to a neuron constituting this neural network comprises a MOS analog circuit whose input/output variables are voltages, and a systollic array is constituted of a large number of these MOS analog circuits.
Inventors:
YONEDA HIDEKI (JP)
SANCHEZ-SINENCIO EDGAR (JP)
SANCHEZ-SINENCIO EDGAR (JP)
Application Number:
PCT/JP1991/001421
Publication Date:
April 29, 1993
Filing Date:
October 17, 1991
Export Citation:
Assignee:
KAWASAKI STEEL CO (JP)
TEXAS A & M UNIV SYS (JP)
TEXAS A & M UNIV SYS (JP)
International Classes:
G06F15/18; G06K9/46; G06N3/063; (IPC1-7): G06G7/60
Foreign References:
| JPH0373080A | 1991-03-28 | |||
| JPH0380379A | 1991-04-05 |
Download PDF:
Previous Patent: PROCESS AND CIRCUITRY FOR SUPPRESSING CONCEALED PICTURE ELEMENTS
Next Patent: ELECTRO-OPTICAL SCANNING SYSTEM
Next Patent: ELECTRO-OPTICAL SCANNING SYSTEM