no code implementations • 25 Sep 2016 • Ahmed M. Abdelsalam, J. M. Pierre Langlois, F. Cheriet
We show that a high accuracy approximation can be necessary in order to maintain the same DNN training and testing performances realized by the exact function.