Hacker News new | ask | show | jobs
by jwuphysics 578 days ago
Could you say a bit more about how so?
1 comments

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.