Kernel size

[Antonio Paes]

I'm with a doubt respect to the kernel values of each conv layers.

I know which has a lot of possibilities to initialize the values of the kernels (gaussian, finetuning from another trained models etc..), my question is:

In each backprop iteration that values from kernel size are updated too?

[Przemek D]

Kernel size is a hyperparameter describing "geometry" of your network, and as such is not modified during neither learning nor inference.