DEFT: Exploiting Gradient Norm Difference between Model Layers for Scalable Gradient Sparsification.

Daegun Yoon, Sangyoon Oh

Browse the full ICPP paper archive.