Learning rate
Weight decay
Momentum
Mini-batch size
Inital
Policy
Gamma
10-4
0.9
16
0.002
Inverse
10-4