AI Tools & Models HOTFeaturedAll
arXiv cs.AIScore60

Hidden not Deleted: How Networks Suppress Entangled Features

Summary

The study shows a neural network that stops reporting a feature may simply be hiding it, not forgetting it — a finding with big implications for how shared representations work.

Why this matters

Relevant to this topic

Open original ↗

Source: arXiv cs.AI · arxiv.org

Related items