"Training Deep Neural Networks with 8-bit Floating Point Numbers."

Naigang Wang et al. (2018)
a service of Schloss Dagstuhl - Leibniz Center for Informatics