Title: Some issues on ranking of classifiers in the presence of label noise
Authors: Efoevi Angelo Koudou - IECL CNRS /Universite de Lorraine (France) [presenting]
Bart Lamiroy - LORIA CNRS - Universite de Lorraine (France)
Abstract: Few instances are reviewed on how the problem of evaluating the performance of classifiers in the presence of noisy labels has been addressed in the recent literature. Some interesting attempts to solve this problem has been carried out under assumptions that could appear to be unrealistic. For instance, some authors used an approach based on the fact that the classifier error is independent of that of the labeller. We discuss how this independence assumption could be relaxed.