True, but that does not show that the approximation can always be learned from e... | Hacker News

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

leod on Aug 18, 2017 | parent | context | favorite | on: Neural Nets Can Learn Function Type Signatures fro...

True, but that does not show that the approximation can always be learned from examples, does it?

bmc7505 on Aug 18, 2017 [–]

No, it does not even guarantee an upper bound on the number of hidden units required to approximate the function with a certain accuracy (even if it can be learned). So its usefulness is very limited.

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