"Training opposing directed models using geometric mean matching."

Jörg Bornschein et al. (2015)
a service of Schloss Dagstuhl - Leibniz Center for Informatics