Research map / Vision-language and multimodal / Multimodal LLMs
Anatomy of a Lie: A Multi-Stage Diagnostic Framework for Tracing Hallucinations in Vision-Language Models
Lexiang Xiong, QI LI, Jingwen Ye, Xinchao Wang
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 all three probes in a GMM instead of using a single probe (ablation) | CAD on Idefics2 · POPE adversarial | AUC: 0.75-0.78 -> 0.947 | Improved |
| use unsupervised CAD instead of supervised probe | CAD vs Supervised Probe · POPE adversarial (average) | AUC: 0.791 -> 0.858 | Improved |
| use CAD instead of 10x-cost Semantic Entropy | CAD vs Semantic Entropy · POPE adversarial (average) | AUC: 0.709 -> 0.858 | Improved |
| use CAD instead of supervised probe | CAD vs Supervised Probe · MME Qwen2-VL | AUC: 0.7408 -> 0.7233 | Hurt |
| use CAD instead of supervised probe | CAD vs Supervised Probe · MME (average) | AUC: 0.7509 -> 0.7960 | Improved |
| contaminate calibration set with up to 30% hallucinations (ablation) | CAD on Idefics2 · POPE adversarial | AUC: above 0.91 at 30% contamination | No clear effect |
| contaminate calibration set with hallucinations (ablation) | CAD on Llava-v1.6 · POPE adversarial | AUC | Hurt |
| contaminate calibration set with hallucinations (ablation) | CAD on Qwen2-VL and DeepSeek-VL2 · POPE adversarial | AUC | No clear effect |
| use H_Evi probe on open-ended captioning (ablation) | H_Evi · MS-COCO captions (N=1000) | H_Evi of hallucinated vs non-hallucinated captions: p << 0.001 | Improved |
Closest papers by meaning
- Auditing Meta-Cognitive Hallucinations in Reasoning Large Language Models (NeurIPS 2025 · similarity 0.92)
- MIRAGE: Assessing Hallucination in Multimodal Reasoning Chains of MLLM (NeurIPS 2025 · similarity 0.92)
- Scalable Token-Level Hallucination Detection in Large Language Models (NeurIPS 2026 · similarity 0.91)
- Combating Multimodal LLM Hallucination via Bottom-Up Holistic Reasoning (AAAI 2025 · similarity 0.90)
- LLM-Check: Investigating Detection of Hallucinations in Large Language Models (NeurIPS 2024 · similarity 0.90)
- HaDeMiF: Hallucination Detection and Mitigation in Large Language Models (ICLR 2025 · 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.
