Isn’t the current best practice to train highly over-parametrized models to zero... | Hacker News

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

		agnosticmantis on Oct 7, 2021 \| parent \| context \| favorite \| on: How to train large deep learning models as a start... Isn’t the current best practice to train highly over-parametrized models to zero training error? That’d be a global optima, no? Unless we’re talking about the optima of test error.

aabaker99 on Oct 7, 2021 [–]

If you find a zero in a non negative function, I would call that a global minima, yes.

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