| 摘要 |
PROBLEM TO BE SOLVED: To separate a learning pattern group that is linearly separable, and improve recognition capability (generalization capability) for an unknown pattern.SOLUTION: Learning is conducted on a learning pattern group using set d (d is a positive integer, where d<D) elements that are a portion of a D-dimensional learning pattern (D is a positive integer greater than or equal to 3), a d-dimensional learning pattern is linearly separated, and an identification line or identification plane is fixed. Learning is conducted on a learning pattern group that is not linearly separated using (d+m) elements (m is a positive integer, where d+m≤D) in place of the d elements, a (d+m)-dimensional learning pattern is linearly separated, and an identification plane is fixed. Then, learning, linear separation, and identification plane fixation are performed using (d+m) elements plus m elements added each time until all of learning pattern groups are linearly separated or the number of elements reaches D.SELECTED DRAWING: Figure 2 |
