SVM and prediction of values

Hello,

I'm new to SVM and I've got a basic understanding problem. I know that I
can classify samples into two groups via +1 and -1. When I want e.g. to
implement a access control based on face recognition I show several
hundred pictures and say if the are allowed to enter (+1) or not (-1).
But what happens when I also want to train the SVM to deliver a name for
a face?
To make it more simple, how would I train a SVM to show the result of a
function, e.g.
f(x,y) = x mod 10 + y mod 20
My first approach was to include the result in the feature set. So I
would not have only x,y as input but f(x,y),x,y as input. Then I've
created three training samples from this input:
1 1:f(x,y) 2:x 3:y
-1 1:f(x,y)-1 2:x 3:y
-1 1:f(x,y)+1 2:x 3:y
But this didn't work, even tough I've scaled the data before I've trained
the SVM. The SVM seemed to ignore the first line and predicted always -1.
I've done a lot of googling and found very much documentation that
explaings the mathematical details but none adressed this issue. I'll
hope anyone is able to point me to the right solution.


regards

J. Luger
.