Is there a way to have layers (esp. a Conv) without biases?

[rohitsupekar]

Or is there a clever way to instruct the training to not train the biases but just the weights?

[DhairyaLGandhi]

#873

[bhvieira]

Sure, just don't pass the bias when you compute your gradients. If you a Conv layer A update only A.weight instead.

You could redefine the Conv struct as well, and remove the bias there.

[CarloLucibello]

duplicate of #868