Research map / Architectures and efficiency / RNNs and normalization
Aging with GRACE: Lifelong Model Editing with Discrete Key-Value Adaptors
Thomas Hartvigsen, Swami Sankaranarayanan, Hamid Palangi, Yoon Kim, Marzyeh Ghassemi
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 |
|---|---|---|---|
| Use GRACE instead of the strongest average baseline, FT, for sequential editing | T5 · zsRE | Avg.: .69 -> .82 | Improved |
| Use GRACE instead of FT+Retrain for sequential editing | BERT · SCOTUS | Avg.: .75 -> .82 | Improved |
| Use GRACE instead of Memory for sequential editing | GPT2-XL · Hallucination | TRR: 25.47 -> 15.84 | Improved |
| Use GRACE instead of Memory for sequential editing | GPT2-XL · Hallucination | ERR: 79.30 -> 7.14 | Improved |
| Use GRACE instead of Memory for sequential editing | GPT2-XL · Hallucination | ARR: 10.07 -> 10.00 | Improved |
| Use GRACE instead of FT for edit time | GPT2-XL · Hallucination | time (s): .26 (.07) -> .13 (.02) | Improved |
| Increase epsilon_init (ablation) | GRACE on T5 · zsRE holdout rephrasings | F1 | Improved |
| Increase epsilon_init (ablation) | GRACE on T5 · zsRE | TRR | Hurt |
| Edit an interior block rather than Block 6 (ablation) | GRACE on T5 · zsRE | ERR | Improved |
| Replace only the last input token rather than all input tokens (ablation) | GRACE on T5 Block 2 · zsRE | TRR | Improved |
| Use warm value initialization instead of cold initialization (ablation) | GRACE on T5 Block 7 · zsRE | edit success | Mixed |
| Use warm rather than cold value initialization (ablation) | GRACE on T5 Block 2 · zsRE | edit success | No clear effect |
Closest papers by meaning
- Towards Lifelong Model Editing via Simulating Ideal Editor (ICML 2025 · similarity 0.91)
- ELDER: Enhancing Lifelong Model Editing with Mixture-of-LoRA (AAAI 2025 · similarity 0.90)
- MEMOIR: Lifelong Model Editing with Minimal Overwrite and Informed Retention for LLMs (NeurIPS 2025 · similarity 0.89)
- Reinforced Lifelong Editing for Language Models (ICML 2025 · similarity 0.88)
- Rethinking Residual Distribution in Locate-then-Edit Model Editing (NeurIPS 2025 · similarity 0.88)
- Lifelong Knowledge Editing for Vision Language Models with Low-Rank Mixture-of-Experts (CVPR 2025 · 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.
