On Derivation of MLP Backpropagation from the Kelley-Bryson Optimal-Control Gradient Formula and Its Application