Describir: On improving generalization in a class of learning problems with the method of small parameters for weakly-controlled optimal gradient systems