TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Good resources on double descent phenomenon?

1 pointsby 0d9eoooalmost 5 years ago
Can anyone recommend any good resources on double descent (DD; the phenomenon whereby overparameterized estimators achieve better error reduction than parameterized ones)?<p>DD seems like voodoo to me. Research has shown that DL models can reproduce even random noise training data perfectly. So DD seems like an assertion that somehow an overparameterized model can contain more information about a source than a sample dataset, like &quot;free information.&quot; It seems like it violates the data processing inequality or something.<p>Is there just confusion about this phenomenon in general, or are there good resources on this?

no comments

no comments