Gradient Descent Training of Bayesian Networks