CONF Mayo-More96a/IDIAP On the Decomposition of Polychotomies into Dichotomies Mayoraz, Eddy Moreira, Miguel EXTERNAL https://publications.idiap.ch/attachments/reports/1996/rr96-08.pdf PUBLIC https://publications.idiap.ch/index.php/publications/showcite/mayo-more96 Related documents Proceedings of The Fourteenth International Conference on Machine Learning 1997 Morgan Kaufmann IDIAP-RR 96-08 REPORT Mayo-More96/IDIAP On the Decomposition of Polychotomies into Dichotomies Mayoraz, Eddy Moreira, Miguel EXTERNAL https://publications.idiap.ch/attachments/reports/1996/rr96-08.pdf PUBLIC Idiap-RR-08-1996 1996 IDIAP Published in the Proceedings of the 14th ICML 1997 Many important classification problems are \emph{polychotomies}, \emph{i.e.} the data are organized into $K$ classes with $K>2$. Given an unknown function $F : Ømega \to \{1, \dots, K\}$ representing a polychotomy, an algorithm aimed at ``learning'' this polychotomy will produce an approximation of $F$, based on a set of pairs $\{(\mathbf{x}^p, F(\mathbf{x}^p))\}_{p=1}^P$. Although in the wide variety of learning tools, there exist some learning algorithms capable of handling polychotomies, many of the interesting tools were designed by nature for dichotomies ($K=2$). Therefore, many researchers are compelled to use techniques to decompose a polychotomy into a series of dichotomies and thus to apply their favorite algorithms to the resolution of a general problem. A decomposition method based on error-correcting codes has been lately proposed and shown to be very efficient. However, this decomposition is designed only on the basis of $K$ without taking the data into account. In this paper, we explore alternatives to this method, still based on the fruitful idea of error-correcting codes, but where the decomposition is inspired by the data at hand. The efficiency of this approach, both for the simplicity of the model and for the generalization, is illustrated by some numerical experiments.