(no subject)

[am]

S. McCandlish, et al. (2018)
"An Empirical Model of
"Large-Batch Training."

"The gradient noise scale
predicts the paralleliz-
ability of neural network
training on a wide range
of tasks."