New Machine Unlearning Method Uses Shrink-and-Heal Lookahead to Remove Data From Language Models
A new arXiv paper proposes unlearning via extrapolating a shrink-and-heal lookahead, aiming to make LLMs behave as if retrained without a forget set, addressing limits of likelihood-suppression objectives.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-03 05:00 · arXiv cs.LG
Unlearning by Extrapolating a Shrink-and-Heal Lookahead