Learning rate

Weight decay

Momentum

Mini-batch size

Inital

Policy

Gamma

10-4

0.9

16

0.002

Inverse

10-4