Hacker News Reader

The point on interpretability is scientific applications is in symbolic regression - MLPs cannot always spit out an equation for some data set: KANs can.

View on HN

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

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

Reply on news.ycombinator.com