Advanced search


Daniel Keysers, Roberto Paredes, Enrique Vidal, Hermann Ney. Comparison of Log-Linear Models and Weighted Dissimilarity Measures. IbPRIA 2003, 1st Iberian Conference on Pattern Recognition and Image Analysis, 2003. Francisco J. Perales (Editors). Lecture Notes in Computer Science, Springer-Verlag.

We compare two successfull discriminative classification algorithms on three databases from the UCI/STATLOG repository. The two approaches are the maximum entropy principle to estimate class posterior probabilitis on the one hand and class-dependent weighted dissimilarity measures for nearest neighbor classifiers on the other hand. The experiments show that the maximum entropy based classifier performs better for the equivalent of a single prototype while the weighted dissimilarity measures performs better for multiple prototypes per class. This interesting result suggests an extension of the maximum entropy method to multiple prototypes.