Dissimilarity Based Vector Space Embedding of Graphs Using Prototype Reduction Schemes

  • Authors:

  • Kaspar Riesen;Horst Bunke

  • Affiliations:

  • Institute of Computer Science and Applied Mathematics, University of Bern, Bern, Switzerland CH-3012;Institute of Computer Science and Applied Mathematics, University of Bern, Bern, Switzerland CH-3012

  • Venue:
  • MLDM '09 Proceedings of the 6th International Conference on Machine Learning and Data Mining in Pattern Recognition
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Graphs provide us with a powerful and flexible representation formalism for object classification. The vast majority of classification algorithms, however, rely on vectorial data descriptions and cannot directly be applied to graphs. In the present paper a dissimilarity representation for graphs is used in order to explicitly transform graphs into n -dimensional vectors. This embedding aims at bridging the gap between the high representational power of graphs and the large amount of classification algorithms available for feature vectors. The basic idea is to regard the dissimilarities to n predefined prototype graphs as features. In contrast to previous works, the prototypes and in particular their number are defined by prototype reduction schemes originally developed for nearest neighbor classifiers. These reduction schemes enable us to omit the cumbersome validation of the embedding space dimensionality. With several experimental results we prove the robustness and flexibility of our new method and show the advantages of graph embedding based on prototypes gained by these reduction strategies.