[jwuphysics]

Could you say a bit more about how so?

[pkoird]

KANs have learnable activations based on splines parameterized on few variables. You can specify a prior over those variables, effectively establishing a prior over your activation function.