The results suggest that
the proposed Bayesian-LSSVM has stronger generalization
ability than does BP-ANN in small sample data modeling. The
weakness of conventional ERM, on which BP-ANN is based,
was exposed. Values generated using a gradient descent algorithm fell easily into local optimum. However, the proposed
model was equivalent to solving convex quadratic programming with linear constraints, and the solution was unique and
global optimum.
منظور از generalization و gradient descent algorithm fell easily into local optimum چیه؟
ممنونم
ability
|