Could you say a bit more about how so? | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		jwuphysics 8 months ago \| parent \| context \| favorite \| on: Bayesian Neural Networks Could you say a bit more about how so?

pkoird 8 months ago [–]

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.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact