Efficiently inaccurate approximation of hyperbolic tangent used as transfer function in artificial neural networks

We propose the approximation of tanh (i.e. the hyperbolic tangent) by specific formation of cubic splines. Thus, we save many multiplications and a division required for the standard double precision evaluation of this function. The cost we have to pay is to admit at most 2-4 decimal digits of accuracy in the final approximation. As a result, a speeding in neural networks performance is experienced after implementing this new approximant as transfer function.