From: A parallel and distributed stochastic gradient descent implementation using commodity clusters
Training time
Groups
16
71.06
a
8
36.74
b
1
22.82
c
4
22.78
2
21.98