Enregistré dans:
Détails bibliographiques
Auteurs principaux: Holmov, Ali, Youssef, Paul, Schoots, Nandi, Seifert, Christin
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:https://arxiv.org/abs/2605.28839
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866911725315424256
author Holmov, Ali
Youssef, Paul
Schoots, Nandi
Seifert, Christin
author_facet Holmov, Ali
Youssef, Paul
Schoots, Nandi
Seifert, Christin
contents Knowledge editing methods such as ROME and MEMIT update factual associations in transformer models by modifying MLP weights. While evaluated mainly by output behavior, their internal mechanism remains underexplored. We investigate whether edits rely on a common mechanism, regardless of which fact is modified. Despite fact-specific weight changes, we argue that ROME and MEMIT target the same subset of weights critical for maintaining edits. To isolate this subset, we train a compact binary mask over the edited weights. The mask reverses 80% of edits on the training set and over 70% on the test set, confirming that diverse edits share a common functional structure. Our analysis reveals that the mask reverses edits by eliminating overattention in later layers. Additionally, we show that injecting the mask during editing drops editing success from 98% to 38%, demonstrating that this mechanism is necessary for edits to succeed. Our finding that edits suppress rather than overwrite knowledge explains why ROME and MEMIT fail to propagate changes to related facts. The identified common functional subspace informs detection and defense against unwanted edits.
format Preprint
id arxiv_https___arxiv_org_abs_2605_28839
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle One Mask to Rule Them All: On Hidden Facts after Editing and How to Find Them
Holmov, Ali
Youssef, Paul
Schoots, Nandi
Seifert, Christin
Machine Learning
Knowledge editing methods such as ROME and MEMIT update factual associations in transformer models by modifying MLP weights. While evaluated mainly by output behavior, their internal mechanism remains underexplored. We investigate whether edits rely on a common mechanism, regardless of which fact is modified. Despite fact-specific weight changes, we argue that ROME and MEMIT target the same subset of weights critical for maintaining edits. To isolate this subset, we train a compact binary mask over the edited weights. The mask reverses 80% of edits on the training set and over 70% on the test set, confirming that diverse edits share a common functional structure. Our analysis reveals that the mask reverses edits by eliminating overattention in later layers. Additionally, we show that injecting the mask during editing drops editing success from 98% to 38%, demonstrating that this mechanism is necessary for edits to succeed. Our finding that edits suppress rather than overwrite knowledge explains why ROME and MEMIT fail to propagate changes to related facts. The identified common functional subspace informs detection and defense against unwanted edits.
title One Mask to Rule Them All: On Hidden Facts after Editing and How to Find Them
topic Machine Learning
url https://arxiv.org/abs/2605.28839