Automatic Kernel Selection: Theory and Algorithms

Kernel methods have found applications throughout much of machine
learning and have been successful in a variety of tasks.  The proper
selection of a kernel function plays a crucial role in the performance
of kernel based algorithms.  However, despite the popularity and
success of kernel-based algorithms, until recently, there has not been
much focus on how to select a kernel function in an automated fashion.
This talk will give an introduction to automatic kernel selection,
covering both theoretical and empirical results.  In particular, I
will present generalization guarantees for this setting, as well as
algorithms for learning weighted combinations of kernels.  This is
joint work with Corinna Cortes and Mehryar Mohri.