New algorithms for multi-class classification with abstention.
problem Learning with abstention in multi-class classification.
method Introducing new surrogate losses and analyzing predictor-rejector framework.
result Strong consistency guarantees for abstention loss function.