it is generally believed that “the exact details of the sigmoid are not critical in a backpropagation network.’’ (p.19, Caudill, 1990). While this is true for the performance of an alreadytrained network, there are important differences that can impact training dramatically,