Research map / Large language models / Retrieval and knowledge
Disentangling Knowledge Representations for Large Language Model Editing
Mengqi Zhang, Zisheng Zhou, Xiaotian Ye, Qiang Liu, Zhaochun Ren, Zhumin Chen, Pengjie Ren
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add disentanglement-based editing (DiKE) to the base model editing pipeline | LLaMA-3 vs AlphaEdit · FINE-KED | Relational Locality avg: 65.6 -> 70.6 | Improved |
| DiKE versus MEMIT-C | GPT-J · FINE-KED | Relational Locality avg: 65.9 -> 71.3 | Improved |
| DiKE versus ROME-C | GPT2-XL · FINE-KED | Relational Locality avg: 50.8 -> 55.0 | Improved |
| DiKE versus AlphaEdit on efficacy | LLaMA-3 · FINE-KED | Efficacy: 98.2 -> 99.1 | No clear effect |
| DiKE versus MEMIT | LLaMA3 (8B) · CounterFact | Avg.: 91.1 -> 92.4 | Improved |
| DiKE versus AlphaEdit | GPT-J (6B) · CounterFact | Avg.: 91.0 -> 90.8 | No clear effect |
| DiKE versus fine-tuning (FT) | GPT-J (6B) · CounterFact | Neighborhood score: 9.8 -> 79.3 | Improved |
| add target-knowledge editing on disentangled representations (TKE) (ablation) | DiKE · FINE-KED, LLaMA3 | Relational Locality | Improved |
| add contrastive disentangling loss (CTR) and knowledge constraint (KC) (ablation) | DiKE KRD module · FINE-KED, LLaMA3 | Relational Locality | Improved |
| add fine-grained irrelevant knowledge constraint (FIK) (ablation) | DiKE DKE module · FINE-KED, LLaMA3 | Relational Locality | Improved |
| increase batch size under subject-consistent batch editing | ROME · FINE-KED | Relational Locality | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- AlphaEdit — AlphaEdit: Null-Space Constrained Knowledge Editing for Language Models (ICLR 2025) · better on 5 of 5 shared results
- MEND — MEND: Meta Demonstration Distillation for Efficient and Effective In-Context Learning (ICLR 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AlphaEdit: Null-Space Constrained Knowledge Editing for Language Models (ICLR 2025 · similarity 0.90)
- Editing Language Model-Based Knowledge Graph Embeddings (AAAI 2024 · similarity 0.89)
- Resolving Lexical Bias in Model Editing (ICML 2025 · similarity 0.89)
- Uncovering Overfitting in Large Language Model Editing (ICLR 2025 · similarity 0.89)
- Editing Memories Through Few Targeted Neurons (AAAI 2025 · similarity 0.89)
- UniEdit: A Unified Knowledge Editing Benchmark for Large Language Models (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
