AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

TARS: MinMax Token-Adaptive Preference Strategy for Hallucination Reduction in MLLMs

KEJIA ZHANG, Keda TAO, Zhiming Luo, Chang Liu, Jiasheng Tang, Huan Wang

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Preference tuning against hallucination · 0 citations (OpenAlex)

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 token-level perturbation (TP) (ablation)TARS · AMBERCog: 0.4 vs 2.5 withoutImproved
add cross-modal alignment score (CAS) token selection (ablation)TARS · AMBERHal-Rate: 4.5-point increase in hallucination without itImproved
add spectral preference alignment (SPA) (ablation)TARS · OBJHalCRi: 3.2 -> 3.7 when removedImproved
replace FFT alignment with token-level contrastive alignment (ablation)TARS · AMBERHal-Rate: 13.2% vs. 16.3%Hurt
perturb visually relevant tokens instead of visual-agnostic ones (ablation)DPO · AMBERHal-Rate: 26.4% to 31.6%Hurt
random token perturbation instead of adaptive (ablation)DPO · AMBERHal-Rate: 28.7% vs. 26.4%; TARS 13.2%Mixed
replace DPO with TARS on 4.8k samples vs 5x LLM augmentationLLaVA-v1.5-7B · AMBERHal-Rate: 16.0% vs. 13.2%Improved
combine TARS with 5x LLM augmentationTARS · AMBERHal-Rate: 13.2 -> 12.3Improved
lower perturbation magnitude omega to 1e-4 (ablation)TARS · AMBERHal: 13.2 -> 15.4Hurt
raise perturbation magnitude omega to 1e-2 (ablation)TARS · AMBERHal: 13.2 -> 20.2Hurt
raise spectral weight lambda from 0.01 to 0.20 (ablation)TARS · AMBERHal: 14.8 -> 12.5Improved
raise spectral weight lambda from 0.20 to 1.00 (ablation)TARS · AMBERHal: 12.5 -> 15.1Hurt

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