Browse by topic
Type of publication
Dimensionality Reduction by Minimising Nearest-Neighbour Classification Error. 2010. Centre for Vision, Speech and Signal Processing (CVSSP) Seminar, Guildford, UKThere is a great interest in dimensionality reduction techniques for tackling the problem of high-dimensional pattern classification. This talk will addresses the topic of supervised learning of a linear dimension reduction mapping suitable for classification problems. The proposed optimization procedure is based on minimizing an estimation of the nearest neighbor classifier error probability, and it learns a linear projection base and a small set of prototypes that support the class boundaries. The learned classifier has the property of being very computationally efficient, making the classification much faster than state-of-the-art classifiers, such as SVMs, while having competitive recognition accuracy. Among the interesting properties of the proposed approach is that the target dimensionality is not restricted either by the number of classes or the number of training samples. Furthermore, a PCA preprocessing, which could potentially ignore some useful information, is not required to achieve a good result.