One Mask to Rule Them All: On Hidden Facts after Editing and How to Find Them

• URL: http://arxiv.org/abs/2605.28839v1
• Date: Sat, 18 Apr 2026 16:18:08 GMT
• Title: One Mask to Rule Them All: On Hidden Facts after Editing and How to Find Them
• Abstract summary: ROME and MEMIT update factual associations in transformer models by modifying weights.<n>We investigate whether edits rely on a common mechanism, regardless of which fact is modified.<n>Our finding that edits suppress rather than overwrite knowledge explains why ROME and MEMIT fail to propagate changes to related facts.
• Score: 7.195629836132288
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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.

Related papers

• Selective Knowledge Edit Reversal via Gated Singular Vector Shrinkage [29.783345718823487]
  We study selective reversal of edited knowledge, where the goal is to reverse targeted edited facts while preserving the remaining edited facts.<n>Based on the hypothesis that each edit is sparsely encoded within the dominant subspace of the edited matrix, we propose a spectral-based reversal framework.<n> Experiments across multiple settings demonstrate the effectiveness of our method in reversing selected edits while preserving unrelated edited facts.
  arXiv  Detail & Related papers  (2026-09-02T04:32:07Z)
• Edit Where You Mean: Region-Aware Adapter Injection for Mask-Free Local Image Editing [28.295031615161136]
  Large diffusion transformers (DiTs) follow global editing instructions well but consistently leak local edits into unrelated regions.<n>We introduce AdaptEdit, a co-trained, instruction- and region-aware adapter framework.<n> AdaptEdit achieves state-of-the-art results, simultaneously outperforming mask-free and oracle-mask baselines.
  arXiv  Detail & Related papers  (2026-04-26T15:28:02Z)
• The Anatomy of an Edit: Mechanism-Guided Activation Steering for Knowledge Editing [52.315369633116255]
  Large language models (LLMs) are increasingly used as knowledge bases, but keeping them up to date requires targeted knowledge editing (KE)<n>We take a mechanistic view of KE using neuron-level knowledge attribution (NLKA)<n>Across representative KE methods, we find a consistent pattern: mid-to-late attention predominantly promotes the new target, while attention and FFN modules cooperate to suppress the original fact.<n>Motivated by these findings, we propose MEGA, a MEchanism-Guided Activation steering method that performs attention-residual interventions in attribution-aligned regions.
  arXiv  Detail & Related papers  (2026-03-21T12:40:15Z)
• Multiplicative Orthogonal Sequential Editing for Language Models [55.42748430481554]
  We propose a new knowledge editing paradigm termed Multiplicative Orthogonal Sequential Editing (MOSE)<n>Compared to current methods, MOSE achieves a 12.08% improvement in sequential editing performance, while retaining 95.73% of general abilities across downstream tasks.
  arXiv  Detail & Related papers  (2026-01-11T04:09:32Z)
• Quantifying Edits Decay in Fine-tuned LLMs [17.377278510871843]
  This study investigates how fine-tuning affects knowledge editing.<n>We evaluate two state-of-the-art editing methods (MEMIT, AlphaEdit) and three fine-tuning approaches.<n>Our results show that edits decay after fine-tuning, with survival varying across configurations.
  arXiv  Detail & Related papers  (2025-11-08T04:58:03Z)
• MEMOIR: Lifelong Model Editing with Minimal Overwrite and Informed Retention for LLMs [76.28901550926021]
  Existing methods for lifelong model editing compromise generalization, interfere with past edits, or fail to scale to long editing sequences.<n>We propose MEMOIR, a novel scalable framework that injects knowledge through a residual memory, while preserving the core capabilities of the pre-trained model.<n>MeMOIR achieves state-of-the-art performance across reliability, generalization, and locality metrics, scaling to thousands of sequential edits with minimal forgetting.
  arXiv  Detail & Related papers  (2025-06-09T16:16:42Z)
• Tracing and Reversing Rank-One Model Edits [5.260519479124422]
  This work investigates the traceability and reversibility of knowledge edits, focusing on the widely used Rank-One Model Editing (ROME) method.<n>We show that ROME introduces distinctive distributional patterns in the edited weight matrices, which can serve as effective signals for locating the edited weights.<n>We propose a method to infer the edited object entity directly from the modified weights, without access to the editing prompt, achieving over 95% accuracy.
  arXiv  Detail & Related papers  (2025-05-27T07:27:01Z)
• MEMIT-Merge: Addressing MEMIT's Key-Value Conflicts in Same-Subject Batch Editing for LLMs [37.374258713584496]
  We show that MEMIT's editing efficacy significantly deteriorates when processing batches containing multiple edits sharing the same subject.<n>We propose MEMIT-Merge, an enhanced approach that merges value processes for facts sharing the same subject.
  arXiv  Detail & Related papers  (2025-02-11T07:42:09Z)
• Related Knowledge Perturbation Matters: Rethinking Multiple Pieces of Knowledge Editing in Same-Subject [49.559994791305535]
  Current state-of-the-art editing methods struggle when tasked with editing multiple related knowledge pieces for the same subject.<n>We introduce the $textS2textRKE$(Same-Subject Related Knowledge Editing) benchmark.<n>Our experiments reveal that only mainstream locate-then-edit methods, such as ROME and MEMIT, exhibit "related knowledge perturbation"
  arXiv  Detail & Related papers  (2025-02-08T04:47:17Z)
• AlphaEdit: Null-Space Constrained Knowledge Editing for Language Models [63.209935157623946]
  Large language models (LLMs) often exhibit hallucinations due to incorrect or outdated knowledge.<n>We introduce AlphaEdit, a novel solution that projects perturbation onto the null space of the preserved knowledge before applying it to the parameters.<n>We theoretically prove that this projection ensures the output of post-edited LLMs remains unchanged when queried about the preserved knowledge.
  arXiv  Detail & Related papers  (2024-10-03T10:06:27Z)
• Model Editing Harms General Abilities of Large Language Models: Regularization to the Rescue [122.20016030723043]
  We evaluate the side effects of model editing on large language models (LLMs) Our analysis reveals that the side effects are caused by model editing altering the original model weights excessively. To mitigate this, a method named RECT is proposed to regularize the edit update weights.
  arXiv  Detail & Related papers  (2024-01-09T18:03:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.