AI Paper Atlas

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

CaliDist: Calibrating Large Language Models via Behavioral Robustness to Distraction

Mohammad Anas Jawad, Cornelia Caragea

ICML 2026 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
apply CaliDist (As.) distraction-based scaling to vanilla logit confidenceLlama-3.1-8B · 7-dataset averageECE: 15.18 -> 4.37Improved
apply CaliDist (As.) to vanilla confidenceQwen3-8B · 7-dataset averageECE: 26.26 -> 3.77Improved
replace temperature scaling with CaliDist (Pr.)Llama-3.1-8B · 7-dataset averageECE: 7.42 -> 4.08Improved
replace self-consistency (15 passes) with CaliDist (As.)Llama-3.1-8B · 7-dataset averageECE: 18.65 -> 4.37Improved
apply CaliDist to verbalized confidenceLlama-3.1-8B · 5-dataset averageECE: 18.00 -> 9.50Improved
apply CaliDist to verbalized confidenceGemini-2.0-Flash · 3-dataset averageECE: 12.69 -> 13.00No clear effect
apply CaliDist to log-prob confidenceGPT-4o-Mini · 3-dataset averageECE: 20.60 -> 2.82Improved
tune sigmoid alpha and beta by grid search instead of default alpha=0, beta=1 (ablation)CaliDist · Llama-3.1 and Qwen-3ECEImproved
use CaliDist instead of SPUQLlama-3.1-8B · CSQAECE: 12.68 -> 4.73Improved
use CaliDist instead of Vector Scaling with 200 validation samplesLlama-3.1-8B · MNLIECE: 3.47 -> 2.80Improved
use CaliDist instead of FSD consistencyLlama-3.1 ContractNLI · ContractNLIECE: 19.16 -> 4.04Improved
use CaliDist (As.) instead of vanilla confidenceQwen3-8B · CSQAAUROC: 70.01 -> 77.62Improved

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