From: ted.dunning@gmail.com   
      
   On May 20, 6:14 pm, "F. Frisch"  wrote:   
   > i have a pool of 8000 koordinates (x1,x2,t) of a 3D-function.   
   >   
   > Has anyone an idee, how i can approximate this funktion with a neuronal   
   > netzwork?   
   >   
   > thank you for all information   
      
   Your problem is not quite well enough defined to have a solution yet.   
      
   Is your function of the form t = f(x1, x2)?  If so, then you can   
   simply scale your output to fit into the normal range of a neural net   
   (typically -1 to 1, but you should use a somewhat smaller range) and   
   then you can train a neural net.  Depending on the software you use,   
   it may help to normalize your inputs somewhat.  A common normalization   
   is to subtract the observed mean and divide by the observed standard   
   deviation.   
      
   There are many other kinds of regression that can deal with this input   
   as well.  There is no reason that neural networks will do better than   
   other approaches, and it definitely will provide you less information   
   in some respects, particularly to do with error estimates.   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)