. . . "However, from the point of learning progress, the weight W is the most preferable for the parameter." . .