AdaPop: Adaptive Popularity for LLM Unlearning Reduces Leakage

Anna Borisiuk · hf · 2026-08-20

Introduces AdaPop (Adaptive Popularity), a method to improve LLM machine unlearning. It adapts gradient pressure based on factor popularity and automates the balance between forgetting and retaining content, aiming to reduce the leakage of unlearned data.

Original post →

More from Research

Research channel →