Batch normalization (gamma and beta parameters)

[Jamal]

Hi guys,

I wanted to ask about the learnable parameters gamma and beta in batch-normalization. If I want to strip away the beta parameters, what argument should I pass?

without learnable parameters: nn.BatchNormalization(N [,eps] [,momentum], false)

Here,passing false argument strips away both  gamma and beta parameter.

[Joe Jacobs]

If you want to disable affine transform (gamma and beta) without specifying eps and momentum, do this:

nn.BatchNormalization(N,nil,nil,false)

where N is the size of the layer.

Joe

[Joe Jacobs]

Sorry just realised I didn't answer your question. Don't think you can just disable the bias without disabling gamma as well.

Joe