To improve learning/recognition efficiency and recognizing ability by detecting a change pattern of a learning/recognition object from difference information of the feature amount of the learning/recognition object and setting the proper number of systems integrated at a learning/recognition time from the change pattern.
A time sequence continuous data input part of an input means 1 inputs the time sequence continuous data to be learned. Then, a sample- formation/quantization part of a feature amount extraction means 2 performs sample formation and quantization. Then, a difference calculation part of the feature amount extraction means 2 calculates a difference with a component one behind of the system, and inputs its difference system and discrete data system to a positive/negative decision part of a system integration parameter setting means 3 as a setting material of a parameter showing a system integration range, and inputs them to a feature amount local part output part of a system learning means 4 as a feature amount. Then, a learning part of the system learning means 4 learns so that the recognition result by a particular partial recognition part becomes a correct category instructed by a teacher data instruction part.
JP2728120 | ECHO CANCELLER |
WO/2011/140632 | METHOD OF IDENTIFICATION AND DEVICES THEREOF |
JP2739825 | PREDICTING METHOD FOR PROTEIN STEREOSCOPIC STRUCTURE |
IMAGAWA TARO
MARUNO SUSUMU