| 摘要 |
<P>PROBLEM TO BE SOLVED: To realize attribute-learning and -transferring which are capable of performing on-line and additional learning. <P>SOLUTION: An attribute-learning and -transferring system 1 comprises: a feature extraction unit 2; a labeling unit 3; a discriminator generation unit 5 for configuring an attribute discriminator with an Adjusted-SOINN, dividing the attribute discriminator into a plurality of pieces, inputting features of training data as an input pattern into a part specified by labeled attribute information, and generating a node and an edge in the Adjusted-SOINN on the basis of the input pattern; a discriminator holding unit 4; an attribute discrimination unit 6 for, when input data is input, calculating first similarity between the node included in the Adjusted-SOINN and the input pattern by inputting the input pattern in each part of the Adjusted-SOINN, and discriminating an attribute of the input data according to the first similarity; and a class discrimination unit 7 for calculating second similarity by comparing an attribute of the input data with attribute information of a class, and discriminating the class according to the second similarity. <P>COPYRIGHT: (C)2012,JPO&INPIT |
