RELEARNING METHOD FOR SUPPORT VECTOR MACHINE AND DEVICE

摘要

PROBLEM TO BE SOLVED: To provide a general method of a perturbation used for relearning of a support vector machine.

SOLUTION: A relearning method for the support vector machine comprises: steps S1 and S2 which perform an initial learning of the SVM by using initial training samples having known labels; a step 3 which selects addition object samples from an initial learning result, and newly generates additional training samples by performing perturbation processing to each addition object sample based on its neighborhood sample; and a step S4 which relearns the SVM by using the initial learning sample and the additional training sample. The perturbation processing is a processing that separates the additional object sample a determined proportion apart from a center of gravity on a straight line connecting the center of gravity of the neighborhood sample and each additional object sample. The additional object sample can be limited to a support vector and whatnot which is discriminated from the initial learning result.

COPYRIGHT: (C)2012,JPO&INPIT