TEACHER SIGNAL GENERATING METHOD FOR POST-PROBABILITY ESTIMATION BY NEURAL NET AND POST-PROBABILITY ESTIMATION SYSTEM FOR NEURAL NET

摘要

PROBLEM TO BE SOLVED: To generate a high-reliability teacher signal even with a little samples when estimating a post-probability through a neural net. ;SOLUTION: A post-probability estimation system is composed of a teacher signal generation processing part 100, learning processing part 200, post- probability estimation processing part 300, input part 400 and output part 500. This teacher signal generation processing part 100 is composed of a distance calculating part 110 for calculating a distance from a concerned learning vector to the other learning vector for each learning vector, near-K selecting part 120 for selecting K pieces of near learning vectors including the concerned vector, class label betting part 130 for respectively generating the teacher signals of learning vectors corresponding to different K values based on the result of betting by betting the class labels of selected learning vectors for each class, and average value calculating part 140 for defining the weight average value of these signals as the teacher signal of the concerned learning vector.;COPYRIGHT: (C)1998,JPO