AdLearn : Un Algorithme d'Apprentissage Interprétable
V. Margot, O. Wintenberger, JP. Baudry, F. Guilloux and C. Geissler. 49èmes Journées de Statistiques, 29 Mai - 2 Juin 2017.
Absract: This presentation introduces a learning algorithm developed in a way that is easily interpretable. This one is based on objects of the form ”If . . . Then . . .”, called experts. The algorithm identifies a set of experts by the minimum contrast method. This set is then aggregated by the use of experts aggregation methods, giving us a predictor whose performance is comparable to that of the best convex combination. The construction of this predictor also allows us to express it as an estimator of the regression function.