Found 1 document(s)
Page 1 of 1
-
We propose a new training setup, called ScaLa, and a new pruning algorithm based on it, called ScaLP. The new training setup ScaLa is designed to make standard Stochastic Gradient Descent resilient to layer width changes...