Leveraging Matrix Computations for Efficient Learning on Graphs

While several kernel functions for graphs have been proposed in the past, their practical applications so far are rather lim- ited. One reason for this could be the fact that graphs are typically large, but applying standard implementations of kernel methods with graph kernels is not the most ecient solution. In this paper, we describe several ways to imple- ment ecient kernel methods for graphs using tools from matrix computations.