Research map / Vision-language and multimodal / Multimodal LLMs
Preference tuning against hallucination
Trains models with preference optimization and targeted data to suppress hallucinated content and locate its causes.
58 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2024–2026) · an approach in Multimodal LLMs, keyword cluster “hallucination · lvlms · mitigating”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
6 papers in 2023–24 and 52 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 2.9% to 2.4%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2024 · Detecting and Preventing Hallucinations in Large Vision Language Models (AAAI 2024 · 145 citations)
- 2025 · Devils in Middle Layers of Large Vision-Language Models: Interpreting, Detecting and Mitigating Object Hallucinations via Attention Lens (CVPR 2025 · 18 citations)
- 2026 · Anatomical Region-Guided Contrastive Decoding: A Plug-and-Play Strategy for Mitigating Hallucinations in Medical VLMs (AAAI 2026 · 1 citation)
Earliest papers here
- Detecting and Preventing Hallucinations in Large Vision Language Models (AAAI 2024 · 145 citations)
- HalluciDoctor: Mitigating Hallucinatory Toxicity in Visual Instruction Data (CVPR 2024 · 48 citations)
- Analyzing and Mitigating Object Hallucination in Large Vision-Language Models (ICLR 2024 · 29 citations)
What has been tried
27 papers here report 337 measured changes (a component added, removed or swapped, compared with a baseline): helped 68%, no clear effect 4%, hurt 26%, mixed 1%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Marking analysis as dispreferred class in FDPO (DA) vs ignoring it | M-HalDetect human eval · Human Eval (fraction truthful) | IA 0.83 vs DA 0.70 at T=0; IA 0.75 vs DA 0.58 at T=1 | Hurt | Detecting and Preventing Hallucinations in Large Vision Language Models (AAAI 2024 · 145 citations) |
| Use consistency threshold 0.5 instead of no HalluciDoctor | hallucination evaluation · average CHAIR | 21.7 -> 13.8 | Helped | HalluciDoctor: Mitigating Hallucinatory Toxicity in Visual Instruction Data (CVPR 2024 · 48 citations) |
| remove co-occurrence factor from revisor training | MSCOCO CHAIR_S · CHAIR_S | 19.7 to 22.6 | Hurt | Analyzing and Mitigating Object Hallucination in Large Vision-Language Models (ICLR 2024 · 29 citations) |
| Apply attention intervention in layers 0-4 instead of layers 5-18 | COCO 2014 val · CHAIRS (CS) | 25.0 -> 31.0 | Hurt | Devils in Middle Layers of Large Vision-Language Models: Interpreting, Detecting and Mitigating Object Hallucinations via Attention Lens (CVPR 2025 · 18 citations) |
| Restrict attention redistribution to visual-perception heads | NoCaps · CIDEr | 76.4 -> 78.8 | Helped | ClearSight: Visual Signal Enhancement for Object Hallucination Mitigation in Multimodal Large Language Models (CVPR 2025 · 14 citations) |
| Add target identification/visual perception and visual perception verification | PhD misleading OR · accuracy | 26.6 -> 35.4 | Helped | Combating Multimodal LLM Hallucination via Bottom-Up Holistic Reasoning (AAAI 2025 · 8 citations) |
| add CBTM | CHAIR-COCO · CHAIR_S | 33.4 -> 18.6 | Helped | Alleviating Hallucinations in Large Vision-Language Models through Hallucination-Induced Optimization (NeurIPS 2024 · 7 citations) |
| add On-Policy Alignment before DPO | AMBER · HalRate | 22.6 -> 11.6 | Helped | Mitigating Hallucinations in Large Vision-Language Models via DPO: On-Policy Data Hold the Key (CVPR 2025 · 6 citations) |
| randomly select among three CD strategies at each generation step | AMBER · CHAIR | 8.0 -> 6.9 | Helped | Octopus: Alleviating Hallucination via Dynamic Contrastive Decoding (CVPR 2025 · 5 citations) |
| add visual-aware token selection | MSCOCO CHAIR · CHAIR_i | 6.43 -> 5.82 | Helped | VASparse: Towards Efficient Visual Hallucination Mitigation via Visual-Aware Token Sparsification (CVPR 2025 · 5 citations) |
| Use DETR and RAM++ together rather than DETR alone | MSCOCO · CHAIR_S | 27.6 -> 17.8 | Helped | Mitigating Object Hallucination in Large Vision-Language Models via Image-Grounded Guidance (ICML 2025 · 4 citations) |
| Add hallucination-aware training | AMBER-G · CHAIR | 7.8 -> 7.2 | Helped | Generate, but Verify: Reducing Hallucination in Vision-Language Models with Retrospective Resampling (NeurIPS 2025 · 2 citations) |
What reviewers will expect
Datasets and baselines named in the 52 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MSCOCO · 12 of 52 papers
- AMBER · 10 of 52 papers
- POPE · 9 of 52 papers
- A-OKVQA · 8 of 52 papers
- GQA · 8 of 52 papers
- MME · 8 of 52 papers
- RLHF-V · 6 of 52 papers
- MMHal-Bench · 5 of 52 papers
Baselines
- VCD · 27 of 52 papers
- OPERA · 19 of 52 papers
- HA-DPO · 10 of 52 papers
- ICD · 9 of 52 papers
- PAI · 9 of 52 papers
- HALVA · 8 of 52 papers
- DoLa · 8 of 52 papers
- POVID · 7 of 52 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 58 papers; from OpenAlex and the papers' affiliations).
- Hong Kong University of Science and Technology · 4 papers
- University of Electronic Science and Technology of China · 3 papers
- ByteDance · 3 papers
- Nanyang Technological University · 2 papers
- National University of Singapore · 2 papers
- Peking University · 2 papers
- Xi’an Jiaotong University · 2 papers
- Rutgers University · 2 papers
- University of North Carolina at Chapel Hill · 2 papers
- University of Michigan - Ann Arbor · 2 papers
Use it
Download BibTeX of 20 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Decoding-time hallucination mitigation (91 papers)
- Bias-aware multi-image robustness (44 papers)
Elsewhere in the topic
- Chain-of-thought multimodal reasoning (332 papers)
- VLM capability analysis (264 papers)
- Multimodal LLM benchmarks (252 papers)
- Spatial reasoning in VLMs (207 papers)
- Instruction-tuned multimodal assistants (199 papers)
- Benchmarks and analysis of MLLMs (181 papers)
- Reinforcement learning for visual reasoning (163 papers)
- Interleaved image-text generation with MLLMs (153 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).