DIDN
DIDN copied to clipboard
Bias in convolutions
Hi,
I have a question regarding the use of bias in the convolutions. I notice that you don't use them (see here for example). What is the rationale behind that? I seem to find no trace of it in the paper.
In our experiment, there was almost no difference in performance depending on the use of bias. So we removed it in the convolution layers as other prior studies did.