AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

S-EDL: Eliciting Self-Evidence from Sequence Likelihoods for Semantic Calibration of LLMs

Yawei Li, Jiazheng Li, David Rügamer, Bernd Bischl, Yulan He, Mina Rezaei

NeurIPS 2026 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes S-EDL, a method that elicits self-evidence from sequence likelihoods to semantically calibrate LLM open-ended generation, where many token sequences share one meaning.

Problem

Calibrating LLMs in open-ended generation is hard because many distinct token sequences express the same meaning, while prior calibration methods target fixed-set classification confidence or binary correctness of individual strings.

Approach

Elicits self-evidence from sequence likelihoods so that confidence is computed at the level of semantic outcomes (meaning clusters) rather than individual strings or fixed label sets; the abstract is truncated so the precise mechanism is not stated.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Fine-tune with S-EDL instead of plain SFTQwen3-8B-Base · Average of 4 datasetssmECE: 16.5 -> 10.5Improved
Fine-tune with S-EDL instead of plain SFTQwen3-8B-Base · Average of 4 datasetsPass@1: 60.9 -> 70.5Improved
Add label smoothing to SFTMiMo-7B-Base · Average of 4 datasetsPass@1: 56.2 -> 53.5Hurt
Add label smoothing to SFTMiMo-7B-Base · Average of 4 datasetssmECE: 12.1 -> 11.9No clear effect
Use binary-reward Dr.GRPO instead of token-level RCFTMinistral3-8B-Base · Average of 4 datasetssmECE: 10.9 -> 11.4Hurt
Use S-EDL instead of Dr.GRPO (cross-domain)Qwen3-8B-Base · MATH-trained, cross-domain averagesmECE: 10.6 -> 5.9Improved
Use S-EDL instead of Dr.GRPO (cross-domain)MiMo-7B-Base · MATH-trained, cross-domain averagesmECE: 12.3 -> 12.2No clear effect
Add LS, RCFT or BLoB to SFT (cross-domain)Qwen3-8B-Base and MiMo-7B-Base · MATH-trained, cross-domain averagesmECEHurt
Vary EDL loss weight over four values (ablation)S-EDL · Ablation of loss weightPass@1, S-Acc, smECEImproved
Raise beta from 1.2 to 1.5 (ablation)S-EDL · Ablation of betasmECEHurt
Set beta to 1.01 (ablation)S-EDL · Ablation of betaPass@1, S-Acc, smECEHurt
Increase N_train from 4 to 32 (ablation)S-EDL · Ablation of N_trainPass@1, S-Acc, smECEImproved

Limitations

Experimental setup

Builds on: Evidential deep learning

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy