in the implicit form (3), where (12)(13) (14)(15) (16) rect substitution of (17) into the network error function (6) is circumvented by means of the adjoined error gradient. The ad- joined error gradient, defined in (8), is derived in terms of the error gradients and , which are easily obtained through classical backpropagation, as shown by the following result.Theorem 2 (Adjoined Gradient): Let (5) define the error func- tion for the network (9), subject to the constraint equation (17) on the network weights. Then, the adjoined gradient is given by is an matrix with columns all equal to , and.
đang được dịch, vui lòng đợi..
