neural networks

Hallo!

I have a modelling problem with my neural network. It seems to me
that the standard NN, like a backpropagation network, uses
transferfunctions that in some way compute output directly from the
inputdata i.e. adding or multiplying stuff or something else.
In my case the targetdata has nothing to do with the inputdata, it is
a set of measurements and not numerically dependant from the input.

How can I train a network like that?
Thank you for the help

Christof Fiedler
.