| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide a discriminator generation device, method and program, and a class recognition unit, method and program, capable of performing learning and recognition at a high speed. <P>SOLUTION: A discriminator generation device 10 comprises: a labeling part 11 for labeling positive or negative information indicating whether or not there is a certain attribute to input data and attaining teacher data; a feature amount extraction part 12 for extracting a feature amount of the teacher data as a weight vector; a feature amount learning part 13 for learning the feature amount extracted by the feature amount extraction part 12 by a self-multiplication type network; and a separation boundary learning part 15 for learning a separation boundary between a positive node and a negative node by an SVM on the basis of the positive or negative information that each weight vector has. The feature amount learning part defines the weight vector as an input node, and determines whether or not to insert a new node on the basis of a distance between the input node and each node. The SVM is made linear and uses a stochastic gradient descent method for estimation of weight. <P>COPYRIGHT: (C)2013,JPO&INPIT |
