I've been reading Rosebrock's "Deep Learning for Computer Vision with Python", and he mentions that in a CNN, one of the layers is a set of $K$ kernels that each activate when they see a specific feature at a specific location.

I realize that this is likely a gross oversimplification, but assuming that the kernels all start out randomized, is there anything preventing many (or all) kernels converging to the same matrix? Without having much knowledge about this, I would guess that it's something with Linear Algebra using some measure to figure out how similar two matrices are and trying to tweak one of them to change?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.