Research map / Large language models / Retrieval and knowledge
Knowledge Editing with Subspace-Aware Key-Value Mappings
Haewon Park, Sangwoo Kim, Yohan Jo
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 |
|---|---|---|---|
| SUIT instead of AlphaEdit | LLaMA3-8B-Instruct · CounterFact batch 100 | Specificity: 31.0 -> 74.2 | Improved |
| SUIT instead of AlphaEdit | GPT-J-6B · CounterFact batch 100 | S: 73.0 -> 84.7 | Improved |
| SUIT instead of AlphaEdit at batch size 1 | LLaMA3-8B-Instruct · CounterFact batch 1 | S: 36.6 -> 85.4 | Improved |
| SUIT instead of AlphaEdit on zsRE | Qwen2.5-7B-Instruct · zsRE Spe. l=1 | Specificity: 30.1 -> 64.4 | Improved |
| SUIT instead of AlphaEdit | LLaMA3-8B-Instruct · General Capability after 1000 edits | GC: 62.2 -> 63.0 | No clear effect |
| add subspace-aware key vector k' only (ablation) | AlphaEdit baseline · CounterFact LLaMA3 | S: 55.8 -> 82.2 | Improved |
| add subspace-aware residual vector delta' only (ablation) | AlphaEdit baseline · CounterFact LLaMA3 | S: 55.8 -> 68.3 | Improved |
| combine k' and delta' (full SUIT) vs k' only (ablation) | SUIT · CounterFact LLaMA3 | S: 82.2 -> 86.8 | Improved |
| add delta' to k' (ablation) | SUIT · CounterFact LLaMA3 | Generalization: 77.9 -> 90.3 | Improved |
| raise energy threshold tau_energy (ablation) | SUIT · CounterFact LLaMA3 | Specificity vs Efficacy | Mixed |
| raise directional penalty lambda (ablation) | SUIT · CounterFact LLaMA3 | Specificity vs Generalization | Mixed |
| project edit onto 2D critical subspace span(w1,w2) vs orthogonal remainder (ablation) | SUIT residual vector · CounterFact 1000 edits | p(a*): 0.59 -> 0.67 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AlphaEdit — AlphaEdit: Null-Space Constrained Knowledge Editing for Language Models (ICLR 2025) · better on 13 of 13 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 0.89)
- SWEA: Updating Factual Knowledge in Large Language Models via Subject Word Embedding Altering (AAAI 2025 · similarity 0.88)
- UniEdit: A Unified Knowledge Editing Benchmark for Large Language Models (NeurIPS 2025 · similarity 0.88)
- Uncovering Overfitting in Large Language Model Editing (ICLR 2025 · similarity 0.88)
- Golden Layers and Where to Find Them: Improved Knowledge Editing for Large Language Models Via Layer Gradient Analysis (NeurIPS 2026 · similarity 0.88)
- Should We Really Edit Language Models? On the Evaluation of Edited Language Models (NeurIPS 2024 · similarity 0.87)
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.
