AI Paper Atlas

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

Enhancing Hallucination Detection through Noise Injection

Litian Liu, Reza Pourreza, Sunny Panchal, Apratim Bhattacharyya, Yubing Jian, Yao Qin, Roland Memisevic

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Data and hallucination detection

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
add MLP-activation noise injection to samplinganswer-entropy detection, Llama-2-7B-chat · GSM8KAUROC: 71.56 -> 76.14Improved
add noise injection across six LLMsanswer-entropy detection · GSM8KAUROC: e.g. Gemma-2B 51.36 -> 57.11; Llama-3.2-3B 76.53 -> 82.70Improved
add noise injectionanswer-entropy detection, Phi-3-mini · TriviaQAAUROC: 82.00 -> 82.02No clear effect
add noise injection without hurting generation accuracyLlama-2-7B-chat · GSM8Kaccuracy: 23.64 -> 24.09No clear effect
add noise injection to five uncertainty metrics (ablation)Predictive entropy, lexical sim., semantic entropy, EigenScore, SelfCheckGPT · TriviaQAAUROC: e.g. 75.70 -> 77.21 (semantic entropy)Improved
inject noise in lower/middle/upper/all layers (ablation)Llama-2-7B-chat · CSQAAUROC: 67.55 -> 69.10 to 70.68Improved
add input perturbation (shuffled ICL examples) on top of noise (ablation)Llama-3.2-3B-Instruct · GSM8KAUROC: 82.70 -> 82.84Improved
add input perturbation alone (ablation)Llama-3.2-3B-Instruct · GSM8KAUROC: 76.53 -> 76.10Hurt
use shared noise across layers rather than independent noise (ablation)Llama-3.2-3B-Instruct · GSM8KAUROC: 80.20 -> 82.70Improved
use positive-mean rather than zero-mean noise (ablation)Llama-3.2-3B-Instruct · CSQAAUROC: 72.27 -> 72.83No clear effect
use uniform instead of bounded Gaussian noise (ablation)Llama-3.2-3B-Instruct · CSQAAUROC: alpha 0.09: 72.64 vs 72.83No clear effect
replace additive noise with inference-time dropout (ablation)Llama-2-7B-chat · CSQAaccuracy: 70.83 -> 67.50Hurt

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