I mean, at the very least I assume that they can generate a function, check that... | Hacker News

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

eximius on Sept 26, 2019 | parent | context | favorite | on: Deep Learning for Symbolic Mathematics

I mean, at the very least I assume that they can generate a function, check that it wasn't in the training set, then solve it and verify.

Then what would be your complaint?

mlevental on Sept 27, 2019 [–]

I don't mean literally memorizing. "memorization" is tongue in cheek for overfit. I'm saying that the net won't actually generalize to classes it hasn't seen examples from (while reduction/transformation algorithms will).

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