Timo Löhr

This paper proposes a principled method to construct credal sets using relative likelihood and ensemble learning, achieving improved uncertainty representation without sacrificing predictive accuracy.