HardTanh function problem

[jacky4323]

Hi,

I have some questions
Why do you use HardTanh function to implement activation function in BNN_cifar100.py?
In other words ,if the previous layer has HardTanh function,and the output of HardTanh function will go through the next layer BinarizedSpatialConvolution which also has binarize function(do the similar thing I think)
I means you will propagate the same gradient two times(HardTanh function and binarize function)

Any help would be appericated! Thanks a lot!

Best Regards,
PengWei