Research map / Large language models / Language, safety and interpretability
Sampling-Free Uncertainty Quantification via Hidden State Dynamics in Language Models
Yixin Bu, Guanyun Zou, Renzhi Wang, Runze Xia, Cunjun Wang, Hongliang Dai, Xiaoqing Ma, Piji Li
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 |
|---|---|---|---|
| combine SSS and CCS instead of using SSS alone (ablation) | Hidden-state UQ · NQ (Llama-8B) | AUROC (%): 70.80 -> 83.67 | Improved |
| combine SSS and CCS instead of using CCS alone (ablation) | Hidden-state UQ · SQuAD (Qwen-7B) | AUROC (%): 49.77 -> 71.85 | Improved |
| use high-entropy attention-head weighted aggregation instead of uniform averaging (ablation) | Hidden-state UQ · NQ (Qwen-7B) | AUROC (%): 68.41 -> 78.41 | Improved |
| use fixed high-entropy head instead of dynamic head selection (ablation) | Hidden-state UQ · NQ (Llama-3B) | AUROC (%): 76.91 vs 76.25 (AVG) and 76.24 (Dynamic) | No clear effect |
| use hidden-state dynamics UQ instead of Semantic Entropy | UQ method · GSM8K (Llama-8B) | AUROC (%): 60.31 -> 71.43 | Improved |
| use hidden-state dynamics UQ instead of SAR | UQ method · NQ (Llama-8B) | PRR (%): 19.38 -> 32.18 | Improved |
| use single-pass hidden-state UQ instead of sampling-based baselines | UQ method · CoQA and NewsQA (Qwen 7B) | execution time: about 25 hours -> about 1 hour | Improved |
| use hidden-state UQ instead of Lexical Similarity | UQ method · CoQA (Llama 1B) | PRR (%): 28.77 -> 21.53 | Hurt |
Closest papers by meaning
- Integrating Local and Global Entropy for Uncertainty Quantification in LLMs (NeurIPS 2026 · similarity 0.91)
- Inv-Entropy: A Fully Probabilistic Framework for Uncertainty Quantification in Language Models (NeurIPS 2025 · similarity 0.91)
- SPANUQ: Span-Level Uncertainty Quantification for Large Language Model Generation (NeurIPS 2026 · similarity 0.90)
- Conformal Prediction with Paraphrase-Aware Scoring for LLM Uncertainty Quantification (NeurIPS 2026 · similarity 0.90)
- Efficient semantic uncertainty quantification in language models via diversity-steered sampling (NeurIPS 2025 · similarity 0.90)
- Distinguishing the Knowable from the Unknowable with Language Models (ICML 2024 · similarity 0.90)
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.
