Kernels on Dynamical Systems and Graphs

by S V N Vishwanathan

We propose a family of kernels based on the Binet-Cauchy theorem, and its extension to Fredholm operators. Our methodology is in the spirit of the behavioral framework invented by Willems, that is, we compare the trajectories of states in order to define kernels. This leads to kernels defined via dynamical systems as well as kernels defined on dynamical systems. Besides their theoretical appeal, we show efficient methods for computing our kernels which make them viable for the practitioner. If time permits, we will also discuss connections with random walk graph kernels.

Joint work with Alex Smola, Rene Vidal, Karsten Borgwardt, and Nic Schraudolph.

Click here to go to the LCI Forum page