INFORMATION PROCESSING SYSTEM, RECOGNITION DICTIONARY LEARNING METHOD, AND RECOGNITION DICTIONARY LEARNING PROGRAM

WIPO Patent Application WO/2014/073366

A1

An information processing system for creating a highly accurate recognition dictionary with regard to an input vector (instance) used in learning. The information processing system comprises: a bag selection means which selects, as an input vector for learning, one bag from a bag group in which a plurality of bags representing a set of instances have been aggregated; an instance selection means which selects one instance from within the selected bag; a distance calculation means which calculates the distance between a plurality of reference vectors and the selected instance; an instance probability calculation means which calculates, on the basis of the calculated distance, an instance correct probability that is the probability that the category of the instance is correct; a bag probability calculation means which calculates, on the basis of the instance correct probabilities pertaining to the plurality of instances included in the bag, a bag correct probability that is the probability that the category of the bag is correct; and a reference vector correction means which corrects a plurality of reference vectors on the basis of the bag correct probability.

HOSOI TOSHINORI (JP)

PCT/JP2013/078567

May 15, 2014

October 22, 2013

Click for automatic bibliography generation  

NEC CORP (JP)

G06N3/00; G06N5/04

TOSHINORI HOSOI: "Object Detection based on Multiple Instance LVQ", IPSJ SIG NOTES 2011 (HEISEI 23) NENDO, 15 October 2011 (2011-10-15), pages 3 - 4

KATO TAKASHI (JP)
Takushi Kato (JP)

View/Download PDF      PDF Help

Previous Patent: CONTROL DEVICE FOR ELECTRIC COMPRESSOR

Next Patent: BACKLASH MEASUREMENT DEVICE AND BACKLASH MEASUREMENT METHOD