1: \begin{abstract} 

2: We present a novel approach to learn a kernel-based

3: regression function. It is based on the use of conical

4: combinations of data-based parameterized kernels and on a new

5: stochastic convex optimization procedure of which we establish

6: convergence guarantees. The overall learning procedure has the nice

7: properties that a) the learned conical combination is automatically

8: designed to perform the regression task at hand and b) the updates

9: implicated by the optimization procedure are quite inexpensive.

10: In order to shed light on the appositeness of our learning strategy,

11: we present empirical results from experiments conducted on

12: various benchmark datasets.

13: \end{abstract}

14: