In convolution, nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=stride, padding=1, bias=False)
What’s the meaning of the bias?
The bias is an additive parameter in the convolution.
It’s like the b
in f(x) = w*x + b
. If you set bias=False
, you will drop the b
term, which might make sense in some cases, e.g. if the next layer is an affine BatchNorm layer.
Each kernel has an own bias term.
However, I think the concept is way better described in Stanford’s CS231n.
5 Likes