Training Deep Neural Networks in 8-bit Fixed Point with Dynamic Shared Exponent Management.

Hisakatsu Yamaguchi, Makiko Ito, Katsuhiro Yoda, Atsushi Ike

Browse the full DATE paper archive.