@inproceedings{,
author={Kocsor, Andr{\'a}s and Kov{\'a}cs, Korn{\'e}l and Szepesv{\'a}ri, Csaba},
title={Margin Maximizing Discriminant Analysis},
abstract={We propose a new feature extraction method called Margin Maximizing Discriminant
Analysis (MMDA) which seeks to extract features suitable for classiZcation
tasks. MMDA is derived based on the principle that an ideal feature should
convey the maximum information about the class labels such that it should
depend only on the geometry of the optimal decision boundary and not on
those parts of the distribution of the input data that do not participate
in shaping this decision boundary. Further, distinct feature components
should convey unrelated information about the data. MMDA satisZes these
principles by projecting input patterns onto the subspace spanned by the
normals of a set of pairwise orthogonal margin maximizing hyperplanes.
Two algorithms are pro- posed to calculate the parameters of the projection
that are shown to yield equivalent results. The kernel mapping idea is
used to derive a corresponding non-linear feature extraction method. Experiments
with several real-world, publicly available data sets demonstrate the competitiveness
of the proposed new method.},
booktitle={Machine Learning: ECML 2004: 15th European Conference on Machine Learning,
vol. 3201},
year={2004},
month={September},
publisher={Springer-Verlag GmbH },
editor={Jean-Fran{\c{c}}ois Boulicaut, Floriana Esposito, Fosca Giannotti, et al.},
pages={227-238}
}