| 摘要 |
PROBLEM TO BE SOLVED: To allow kernel functions to have the sufficient power of expression to an object problem by adaptively updating the kernel function even when the states or circumstances of the objects of kernel functions are different. SOLUTION: In a kernel function updating part 1, an update processing part 11 calculates a parameterβbased on teacher part 3, and decides the kernel functions based on the parameterβ, and when an evaluation reference value is a predetermined threshold or less, the kernel function updating part 1 updates the kernel functions by adding the kernel functions whose parameterβhas been updated. At that time, an evaluation reference value calculating part 12 calculates an evaluation reference value by calculating a gram matrix and its Moore Penn Lowe's general reversing based on the teacher data 3 and the kernel functions. In a kernel function processing part 2, a recognition result calculating means 21 calculates a recognition result based on an input vector 4 of an unknown sample, the teacher data 3, the updated kernel functions and the Moore Penn Lowe's general reversing, and a class label calculating means 22 calculates a class label based on the recognition result. COPYRIGHT: (C)2006,JPO&NCIPI
|
