Research map / Probabilistic and scientific modeling / Molecules and proteins
AnyEdit++: Adaptive Long-Form Knowledge Editing via Bayesian Surprise
Bowen Tian, Caixue He, Jiemin Wu, Jingying Wang, Wenshuo Chen, Zexi Li, Yutao Yue
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 |
|---|---|---|---|
| replace fixed-window chunking with Bayes-Chunk surprisal-based segmentation (ablation) | AnyEdit · EditEverything average | BLEU: 72.64 -> 75.00 (Llama-3.1-8B-Instruct) | Improved |
| replace fixed-window chunking with Bayes-Chunk (ablation) | AnyEdit · EditEverything average | BLEU: 42.30 -> 50.13 (Llama-2-7B) | Improved |
| replace fixed-window chunking with Bayes-Chunk (ablation) | AnyEdit · EditEverything average | BLEU: 81.81 -> 85.33 (Qwen-2.5-7B-Instruct) | Improved |
| replace fixed-window chunking with Bayes-Chunk (ablation) | AnyEdit · EditEverything Code | BLEU: 50.78 -> 70.50 (Llama-2-7B) | Improved |
| replace fixed-window chunking with Bayes-Chunk (ablation) | AnyEdit · EditEverything BERTScore average | BERTScore: 94.23 -> 94.50 (Llama-3.1-8B-Instruct) | Improved |
| replace fixed-window chunking with Bayes-Chunk (ablation) | AnyEdit · UnKE | BLEU: 79.02 -> 81.57 | Improved |
| replace fixed-window chunking with Bayes-Chunk (ablation) | AnyEdit · CounterFact | BLEU: 86.27 -> 90.69 | Improved |
| replace fixed-window chunking with Bayes-Chunk (ablation) | AnyEdit · Poetry (Llama-3.1-8B-Instruct) | BERTScore: 85.80 -> 84.78 | Hurt |
| replace fixed-window chunking with Bayes-Chunk (ablation) | AnyEdit · News (Qwen-2.5-7B-Instruct) | BLEU: 80.09 -> 79.97 | No clear effect |
| apply Bayes-Chunk to longer, more logical texts (ablation) | AnyEdit · QwQ-Edit (Qwen-2.5-7B-Instruct) | BLEU/BERTScore | Improved |
| add Bayes-Chunk segmentation to fine-tuning-based FT-UKE (ablation) | FT-UKE · EditEverything average (Llama-3.1-8B-Instruct) | BLEU: 99.90 -> 99.95 | Improved |
| add Bayes-Chunk segmentation to FT-UKE (ablation) | FT-UKE · Poetry/News (Qwen-2.5-7B-Instruct) | BLEU: Math 99.97 -> 99.79; average 99.52 -> 99.57 | Mixed |
Closest papers by meaning
- AnyEdit: Edit Any Knowledge Encoded in Language Models (ICML 2025 · similarity 0.90)
- AlphaEdit: Null-Space Constrained Knowledge Editing for Language Models (ICLR 2025 · similarity 0.89)
- Uncovering Overfitting in Large Language Model Editing (ICLR 2025 · similarity 0.88)
- Knowledge in Superposition: Unveiling the Failures of Lifelong Knowledge Editing for Large Language Models (AAAI 2025 · similarity 0.88)
- In-Context Editing: Learning Knowledge from Self-Induced Distributions (ICLR 2025 · similarity 0.87)
- UniEdit: A Unified Knowledge Editing Benchmark for Large Language Models (NeurIPS 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.
