Some Approximation Properties of Projection Pursuit Learning Networks