Stochastic low-rank kernel learning for regression

We present a novel approach to learn a kernel-based regression function. It is based on the use of conical combinations of data-based parameter-ized kernels and on a new stochastic convex op-timization procedure of which we establish con-vergence guarantees. The overall learning pro-cedure has the nice properties that a) the learned conical combination is automatically designed to perform the regression task at hand and b) the updates implicated by the optimization proce-dure are quite inexpensive. In order to shed light on the appositeness of our learning strategy, we present empirical results from experiments con-ducted on various benchmark datasets. 1