Given a neural network object, this function returns the closed, symbolic, expression implemented by the network (as a string).
This allows you to use a neural network model without relying on the neural network toolbox.
Note I only implemented for feed forward nets (MLPs) and not all possible transfer functions are supported. However, it should be very straightforward to do this.
>> net = newff([-1 1; -1 1],[3 1]);
(2/(1+exp(-2*((2/(1+exp(-2*(x1*1.728941e+00 + x2*1.700224e+00 + -2.424871e+00)))-1)*-9.045580e-01 + (2/(1+exp(-2*(x1*-2.422662e+00 + x2*-1.034790e-01 + 000000)))-1)*-1.976229e-01 + (2/(1+exp(-2*(x1*2.044171e+00 + x2*1.304364e+00 + 2.424871e+00)))-1)*1.050105e+00 + 000000)))-1)
This function originates from the Surrogate Modeling (SUMO) Toolbox : http://www.sumo.intec.ugent.be