Research map / Vision-language and multimodal / Multimodal LLMs
DARE to Mitigate Hallucination: Dual-path Auto-Regressive-aware Editing
Jaeho Lee, Jeongeun Lee, Gyeong-Moon Park
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 |
|---|---|---|---|
| add AR-aware editing to TF editing (ablation) | DARE (LLaVA-1.5) · CHAIR | CHAIR_S: 20.53 -> 15.83 | Improved |
| add visual-difference editing to AR editing (ablation) | DARE (LLaVA-1.5) · CHAIR | CHAIR_S: 17.61 -> 15.57 | Improved |
| add visual-difference editing to TF editing (ablation) | DARE (LLaVA-1.5) · CHAIR | CHAIR_S: 20.53 -> 17.17 | Improved |
| add AR editing to TF+visual-difference editing (full DARE) (ablation) | DARE (LLaVA-1.5) · CHAIR | CHAIR_S: 17.17 -> 14.93 | Improved |
| add TF editing to AR+visual-difference editing (full DARE) (ablation) | DARE (LLaVA-1.5) · CHAIR | CHAIR_S: 15.57 -> 14.93 | Improved |
| add visual-difference editing to TF+AR editing (full DARE) (ablation) | DARE (LLaVA-1.5) · CHAIR | CHAIR_I: 6.47 -> 6.07 | Improved |
| AR-aware editing alone vs TF-based editing alone (ablation) | LLaVA-1.5 · CHAIR | CHAIR_S: 20.53 -> 17.61 | Improved |
| edit layers 16-31 instead of 0-15 (ablation) | AR-aware editing (LLaVA-1.5) · CHAIR | CHAIR_S: 17.70 -> 15.11 | Improved |
| edit layers 16-31 instead of 24-31 (ablation) | AR-aware editing (LLaVA-1.5) · CHAIR | CHAIR_S: 16.87 -> 15.11 | Improved |
| HalluSpace rank k=2 instead of k=8 (ablation) | AR-aware editing (LLaVA-1.5) · CHAIR | CHAIR_S: 17.09 -> 15.11 | Improved |
| HalluSpace rank k=2 instead of k=1 (ablation) | AR-aware editing (LLaVA-1.5) · CHAIR | CHAIR_S: 18.44 -> 15.11 | Improved |
| temporal window 3 instead of 1 (ablation) | AR-aware editing (LLaVA-1.5) · CHAIR | CHAIR_S: 16.92 -> 15.11 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Nullu — Nullu: Mitigating Object Hallucinations in Large Vision-Language Models via HalluSpace Projection (CVPR 2025) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AFTER: Mitigating the Object Hallucination of LVLM via Adaptive Factual-Guided Activation Editing (ICLR 2026 · similarity 0.93)
- Hallucination-aware Intermediate Representation Editing in Large Vision-Lanugage Models (ICLR 2026 · similarity 0.92)
- Self-Correcting Decoding with Generative Feedback for Mitigating Hallucinations in Large Vision-Language Models (ICLR 2025 · similarity 0.92)
- Look Carefully: Adaptive Visual Reinforcements in Multimodal Large Language Models for Hallucination Mitigation (ICLR 2026 · similarity 0.92)
- Causally-Grounded Dual-Path Attention Intervention for Object Hallucination Mitigation in LVLMs (AAAI 2026 · similarity 0.92)
- Learning to Decode Against Compositional Hallucination in Video Multimodal Large Language Models (ICML 2026 · similarity 0.91)
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.
