The Weisfeiler-Lehman Kernel
Pascal Schweitzer (ANU)
NICTA SML SEMINARDATE: 2011-05-12
TIME: 11:00:00 - 12:00:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
Over the past, kernel methods have gained increasing attention for the use in machine learning tasks. Their popularity stems from wide applicability in other well developed techniques such as support vector machines.
Graph kernels constitute kernel methods that are applied to structured data that is available as relational information, typically modeled by graphs. A major challenge for the research on graph kernels is the design of kernels sufficiently efficient to be applicable in practice in order to allow large scale data analysis. I will describe the Weisfeiler-Lehman Kernel, a highly efficient and thus scalable graph kernel. It is based on graph isomorphism techniques that gather information on subgraphs by color refinement. This enables the kernel to make use of meaningful structure information on the input data at low computational cost.
This is joint work with Nino Shervashidze, Erik Jan van Leeuwen, Kurt
Mehlhorn and Karsten Borgwardt.
