Title :
Linear discriminant analysis and discriminative log-linear modeling
Author :
Keysers, Daniel ; Ney, Hermann
Author_Institution :
Dept. of Comput. Sci., Rheinisch-Westfalische Tech. Hochschule Aaachen Univ., Aachen, Germany
Abstract :
We discuss the relationship between the discriminative training of Gaussian models and the maximum entropy framework for log-linear models. Observing that linear transforms leave the distributions resulting from the log-linear model unchanged, we derive a discriminative linear feature reduction technique from the maximum entropy approach and compare it to the well-known linear discriminant analysis. From experiments on different corpora we observe that the new technique performs better than linear discriminant analysis if the dimensionality of the feature space is large with respect to the number of classes.
Keywords :
Gaussian distribution; maximum entropy methods; pattern recognition; Gaussian models; discriminative linear feature reduction technique; discriminative log-linear modeling; discriminative training; linear discriminant analysis; linear transforms; maximum entropy framework; Character generation; Computer science; Context modeling; Contracts; Entropy; Linear discriminant analysis; Maximum likelihood estimation; Pattern recognition; Thermodynamics; Training data;
Conference_Titel :
Pattern Recognition, 2004. ICPR 2004. Proceedings of the 17th International Conference on
Print_ISBN :
0-7695-2128-2
DOI :
10.1109/ICPR.2004.1334033