Section: New Results

Statistical learning and mathematical modeling interactions

Participants : Damiano Lombardi, Fabien Raphel.

In [30] a greedy dimension reduction method is proposed to deal with classification problems. The method proposed can be seen as a goal oriented dimension reduction method. Elements of a Stiefel manifold (whose dimension is not fixed a priori) are computed in such a way that the classification score is maximised. Several examples are proposed to illustrate the method features and to highlight its differences with classical reduction methods used in classification.