The point on interpretability is scientific applications is in symbolic regressi... | Hacker News

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

		throwaway2562 10 months ago \| parent \| context \| favorite \| on: Kolmogorov-Arnold networks may make neural network... The point on interpretability is scientific applications is in symbolic regression - MLPs cannot always spit out an equation for some data set: KANs can.

buildbot 10 months ago [–]

I thought that MLPs are universal function approximators? https://en.wikipedia.org/wiki/Universal_approximation_theore...

tomrod 10 months ago | [–]

One of the beautiful cosmic injustices about mathematics is that proofs about objects need not be constructive to prove existence.

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