In contrast to the standard learning paradigm where all classes can be
observed in training data, learning with augmented classes (LAC) tackles the
problem where augmented classes unobserved in the training data may emerge in
the test phase. Previous research showed that given unlabele