Session 3 Keynote

Abstract

Summary form only given. Recently, an emerging trend of representing objects by graphs can be observed. As a matter of fact, graphs offer a versatile alternative to feature vectors in many subfields of intelligent information processing, including pattern recognition, machine learning, and data mining. However, the space of graphs contains almost no mathematical structure, and consequently, there is a lack of suitable methods for graph classification, clustering and related tasks. In this talk, we propose a general approach to transforming graphs into n-dimensional real vector spaces by means of graph dissimilarity. As a matter of fact, this approach makes all algorithmic tools originally developed for feature vectors instantly available to graphs. In particular, it leads to a novel family of graph kernels. Moreover, the embedding procedure allows one to generate, in a straightforward way, systems for classification and clustering that involve multiple experts. With various experimental results we prove the robustness and flexibility of this new approach and show that it outperforms standard graph classification and clustering methods on several graph data sets of diverse nature.