Research map / Vision-language and multimodal / Multimodal LLMs
ADAPT: Attention Dynamics Alignment with Preference Tuning for Faithful MLLMs
Zhiyuan Yao, Zheren Fu, Zhixiao Zheng, Jiajun Li, Yi Tu, Zhendong Mao
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 Visual Enhance (attention anchor) (ablation) | LLaVA-v1.5-7B · AMBER | Chair: 7.8 -> 4.7 | Improved |
| add attention-supervised inference (ablation) | LLaVA-v1.5-7B · AMBER | Chair: 7.8 -> 6.3 | Improved |
| add VAG-DPO (ablation) | LLaVA-v1.5-7B · AMBER | Chair: 7.8 -> 6.0 | Improved |
| combine all three modules (full ADAPT) vs best single module (ablation) | LLaVA-v1.5-7B · AMBER | Chair: 4.7 -> 3.8 | Improved |
| add VAG-DPO on top of training-free ADAPT-TF (ablation) | LLaVA-v1.5-7B · Obj-Hal | Chair_s: 35.6 -> 30.7 | Improved |
| add VAG-DPO on top of training-free ADAPT-TF (ablation) | LLaVA-v1.5-13B · AMBER | Chair: 4.0 -> 4.2 | Hurt |
| add VAG-DPO on top of ADAPT-TF (ablation) | Qwen2.5-VL-7B · MM-Hal | Score: 3.83 -> 3.75 | Hurt |
| add ADAPT | LLaVA-v1.5-7B · POPE-Adv | Precision: 76.1 -> 91.9 | Improved |
| add ADAPT | Qwen2.5-VL-7B · Obj-Hal | Chair_s: 32.3 -> 16.8 | Improved |
| add ADAPT | Qwen2.5-VL-3B · Obj-Hal | Chair_s: 43.2 -> 15.2 | Improved |
| use ADAPT instead of OPERA decoding | LLaVA-v1.5-7B · Inference time | Time ratio: 7.20 -> 1.42 | Improved |
| use VAG-DPO instead of vanilla DPO (ablation) | LLaVA-v1.5-7B · MME | Score: 1443.20 -> 1459.95 | Improved |
Closest papers by meaning
- Causally-Grounded Dual-Path Attention Intervention for Object Hallucination Mitigation in LVLMs (AAAI 2026 · similarity 0.93)
- Look Carefully: Adaptive Visual Reinforcements in Multimodal Large Language Models for Hallucination Mitigation (ICLR 2026 · similarity 0.93)
- Not All Tokens and Heads Are Equally Important: Dual-Level Attention Intervention for Hallucination Mitigation (AAAI 2026 · similarity 0.93)
- Dynamic Multimodal Activation Steering for Hallucination Mitigation in Large Vision-Language Models (ICLR 2026 · similarity 0.92)
- Mitigating Hallucinations in Large Vision-Language Models via Causal Route Gating (ICML 2026 · similarity 0.92)
- ASCD: Attention-Steerable Contrastive Decoding for Reducing Hallucination in MLLM (AAAI 2026 · similarity 0.92)
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.
