PhD Program Summer School 2006
Graph Mining: Selected methods for learning to classify graph-structered objects
(2 Lectures, Part 1 and Part 2)
| speaker | Prof. Dr. Stefan Wrobel, Fraunhofer Institut, Sankt Augustin |
| date | Part 1: September 27, 2006 Part 2: September 28, 2006 |
| abstract | In many application areas, graphs are a natural formalism for representing the objects under consideration. As a most prominent example, this is the case in pharmacy and chemistry, where molecules can naturally be represented by their chemical graph structure. Since in many domains, large example data bases are available that are difficult to analyse manually, recently there has been increased interest in automated methods of data mining and machine learning that can handle graph-structured objects. In this tutorial we will focus on two different approaches to this problem that we have investigated in our own work. First, approaches using kernel methods and appropriately designed kernel functions for graphs. We will introduce the basic principles of kernel methods and focus on the challenges and solutions of finding kernel functions for them. The second approach is oriented towards frequent patterns. Here, we will give details of one approach that uses a special class of planar graphs to arrive at a frequent pattern method that, in contrast to most of the methods for general graphs, can be shown to be tractable. |