Research map / Large language models / Retrieval and knowledge
Energy-Regularized Sequential Model Editing on Hyperspheres
Qingyuan Liu, Jia-Chen Gu, Yunzhi Yao, Hong Wang, Nanyun Peng
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 SPHERE sparse-space projection to AlphaEdit | AlphaEdit, LLaMA3-8B · CounterFact | efficacy: 4.37 -> 52.89 | Improved |
| add SPHERE projection to AlphaEdit | AlphaEdit, LLaMA3-8B · ZsRE | specificity: 28.78 -> 45.40 | Improved |
| add SPHERE projection to AlphaEdit | AlphaEdit, Qwen2.5-7B · ZsRE | efficacy: 42.01 -> 70.04 | Improved |
| add SPHERE projection to AlphaEdit | AlphaEdit, Qwen2.5-7B · CounterFact | efficacy: 43.92 -> 60.76 | Improved |
| add SPHERE projection to AlphaEdit | AlphaEdit, Qwen2.5-7B · CounterFact | consistency: PRUNE 29.50 vs SPHERE 14.74 | Hurt |
| add SPHERE projection to AlphaEdit | AlphaEdit, LLaMA3-8B · CounterFact | fluency: 482.36 -> 551.51 | Improved |
| insert SPHERE projection into baseline editing methods | FT, MEMIT, RECT, PRUNE, AlphaEdit, LLaMA3-8B · 3000 sequential edits | efficacy: 49.05% | Improved |
| insert SPHERE projection into baseline editing methods | baseline editors, LLaMA3-8B · 3000 sequential edits | generalization: 42.64% | Improved |
| insert SPHERE projection into baseline editing methods | baseline editors, LLaMA3-8B · 3000 sequential edits | specificity: 24.44% | Improved |
| add SPHERE projection | edit methods, LLaMA3-8B · general abilities after 15k edits (GSM8K, RTE, NQ, BoolQ) | general performance | Improved |
| vary cumulative ratio eta and suppression strength alpha of SPHERE (ablation) | MEMIT + SPHERE, LLaMA3-8B · 3000 sequential edits | editing performance | 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 16 of 16 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
- Edit Less, Achieve More: Dynamic Sparse Neuron Masking for Lifelong Knowledge Editing in LLMs (NeurIPS 2025 · similarity 0.89)
- Editing Memories Through Few Targeted Neurons (AAAI 2025 · similarity 0.89)
- MEMOIR: Lifelong Model Editing with Minimal Overwrite and Informed Retention for LLMs (NeurIPS 2025 · similarity 0.89)
- Perturbation-Restrained Sequential Model Editing (ICLR 2025 · similarity 0.89)
- WISE: Rethinking the Knowledge Memory for Lifelong Model Editing of Large Language Models (NeurIPS 2024 · similarity 0.88)
- Precise Localization of Memories: A Fine-grained Neuron-level Knowledge Editing Technique for LLMs (ICLR 2025 · similarity 0.88)
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.
