SCoOP: Semantic Consistent Opinion Pooling for Uncertainty Quantification in Multiple Vision-Language Model Systems

Chung-En Johnny Yu; Brian Jalaian; Nathaniel D. Bastian

cs.AIMar 25, 2026

SCoOP: Semantic Consistent Opinion Pooling for Uncertainty Quantification in Multiple Vision-Language Model Systems

Authors: Chung-En Johnny Yu, Brian Jalaian, Nathaniel D. Bastian

Organizations: University of West Florida Pensacola, Florida, USA · United States Military Academy West Point, New York, USA

Abstract

Combining multiple Vision-Language Models (VLMs) can enhance multimodal reasoning and robustness, but aggregating heterogeneous models' outputs amplifies uncertainty and increases the risk of hallucinations. We propose SCoOP (Semantic-Consistent Opinion Pooling), a training-free uncertainty quantification (UQ) framework for multi-VLM systems through uncertainty-weighted linear opinion pooling. The core idea is to treat each VLM as a probabilistic "expert," sample multiple outputs, map them to a unified space, aggregate their opinions, and produce a system-level uncertainty score. Unlike prior UQ methods designed for single models, SCoOP explicitly measures collective, system-level uncertainty across multiple VLMs, enabling effective hallucination detection and abstention for highly uncertain samples. On ScienceQA, SCoOP achieves an AUROC of 0.866 for hallucination detection, outperforming baselines (0.732-0.757) by approximately 10-13%. For abstention, it attains an AURAC of 0.907, exceeding baselines (0.818-0.840) by 7-9%. Despite these gains, SCoOP introduces only microsecond-level aggregation overhead relative to the baselines, which is trivial compared to typical VLM inference time (on the order of seconds). These results demonstrate that SCoOP provides an efficient and principled mechanism for uncertainty-aware aggregation, advancing the reliability of multimodal AI systems. Our code is publicly available at https://github.com/chungenyu6/SCoOP.

Figures & tables

Figure 1: Overview of SCoOP increasing the reliability of a multi-VLM system.

Figure 2: SCoOP (Semantic-Consistent Opinion Pooling) workflow. For each VLM $M_{k}$ , we sample $N$ responses, map them to the unified option set $\Theta$ , and form a probability vector $\mathbf{p}_{k}$ . Each model’s uncertainty weight $w_{k}$ is calculated from its Shannon entropy. We then apply weighted linear opinion pooling $\mathbf{p}_{\text{agg }}=\sum_{k=1}^{K}w_{k}\mathbf{p}_{k}$ to obtain the system distribution, which can obtain the final response $\theta^{*}$ and the system-level uncertainty $H_{agg}$ . It unifies heterogeneous VLMs’ outputs and quantifies uncertainty of the multi-VLM system.

Figure 3: Hallucination detection (AUROC) and abstention (AURAC) performance on ScienceQA with 3-VLM systems of extra-large-parameter models. Higher values indicate better uncertainty quality and impacts. The green dashed line marks the average AUROC/AURAC of the three individual VLMs evaluated separately (without aggregation).

Table 4

Figure 4: UQ performance across varying model parameter scales. Methods are evaluated in 3-VLM systems on ScienceQA. AUROC for hallucination detection; AURAC for abstention.

Figure 5: UQ performance across varying system sizes on ScienceQA. Methods are evaluated across various 2- to 5-VLM systems using large-scale models. AUROC for hallucination detection; AURAC for abstention.

Table 3: AUROC under conditions where all individual VLMs yield low benchmark accuracy (

<

50%). Each configuration corresponds to

K

-VLM systems with medium scale models. Bold values indicate the highest score for each VLM system. Underline values indicate the second highest score.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Symbol	Description
$K$	Number of VLMs (models or experts) in the system
$M_{k}$	The $k$ -th VLM
$\mathcal{M}=\{M_{1},\ldots,M_{K}\}$	Multi-VLM system composed of $K$ models
$x=(I,q)$	Multiple-choice question (MCQ) input with image $I$ and text question $q$
$N$	Number of sampled responses per model
$y_{k}^{n}$	$n$ -th raw response sampled from model $M_{k}$

Appendix

Table 4: Notation summary for the SCoOP framework.

Panel of Figure 6 — Figure 6: Illustrations of the aggregation baselines.

Model Size	Models	Combination Count
Small (2–4B)	gemma-3-4b-it , Qwen2.5-VL-3B-Instruct , InternVL3-2B-Instruct , deepseek-vl2-tiny	2-VLM: 6 3-VLM: 4 4-VLM: 1
Medium (12–16B)	gemma-3-12b-it , llava-v1.6-vicuna-13b-hf , InternVL3-14B-Instruct , deepseek-vl2-small	2-VLM: 6 3-VLM: 4 4-VLM: 1
Large (27–38B)	gemma-3-27b-it , llava-v1.6-34b-hf , Qwen2.5-VL-32B-Instruct , InternVL3-38B-Instruct , deepseek-vl2	2-VLM: 10 3-VLM: 10 4-VLM: 5 5-VLM: 1
Extra Large (72–78B)	llava-next-72b-hf , Qwen2.5-VL-72B-Instruct , InternVL3-78B-Instruct	2-VLM: 3 3-VLM: 1

Appendix

Table 5: Comprehensive list of VLMs used in our experiments and the number of possible model combinations per system size

K

Panel of Table 6 — Table 6: Prompt template for VLM generating discrete choices.

Hallucination Detection in VLMs

Hallucination detection in vision-language models identifies claims that are unsupported by images, video, or other visual evidence.

See all 72 papers

CUSP: Decomposable Collective Uncertainty for Multi-Agent Multimodal Reasoning
Sep 4, 2026Chung-En Johnny Yu, David Garcia, Brian Jalaian +1Language Model Ensembles Multimodal Reasoning
University of West Florida Pensacola, Florida, USA · United States Military Academy West Point, New York, USA
UHP Detection: LVLMs have their Unique Hallucination Pattern in the Consistency Space
Aug 4, 2026Amir Mohammad Ezzati, Kiyan Rezaee, Bardiya Kariminia +4Hallucination Detection in VLMs Vision-Language Models
Sharif University of Technology · Michigan State University · Shahid Beheshti University +2
The Cost of Reasoning: Chain-of-Thought Induces Overconfidence in Vision-Language Models
Mar 17, 2026Robert Welch, Emir Konuk, Kevin SmithConfidence Estimation for VLMs VLM Reasoning
KTH Royal Institute of Technology, Stockholm, Sweden · Science for Life Laboratory, Stockholm, Sweden

Sep 4, 2026J/K move · Enter open · S save

CUSP: Decomposable Collective Uncertainty for Multi-Agent Multimodal Reasoning

Chung-En Johnny Yu, David Garcia, Brian Jalaian, Nathaniel D. Bastian

University of West Florida Pensacola, Florida, USA · United States Military Academy West Point, New York, USA

Language Model Ensembles Multimodal Reasoning Multimodal Agents

Open paper AskPDF arXiv

Aggregating heterogeneous vision-language models (VLMs) can improve multimodal reasoning, but neither an individual model's confidence nor that of the aggregated answer measures reliability at the system level. We present CUSP (Collective Uncertainty through Semantic Opinion Pooling), a training-free uncertainty quantification framework that maps multiple VLM responses to a shared semantic response space, pools them into a pooled semantic opinion, and reports two complementary system-level signals: collective uncertainty, the dispersion of the pooled opinion, and Jensen-Shannon divergence (JSD), the conflict among the model-level opinions. Within this pooled semantic opinion, the unnormalized collective entropy decomposes exactly into the mean of the models' individual semantic entropies and the JSD, separating total dispersion from model conflict. Requiring neither token logits nor calibration labels, CUSP applies to open-weight and commercial VLMs alike. In static multi-VLM ensembles, collective uncertainty is the strongest signal in the small-model regime (0.764 AUROC for prediction-error detection, 0.889 AUARC for abstention), outperforming uncertainty baselines majority voting and naive selection by 4.7 to 15.8 points and widening its margin as the ensemble grows; JSD is strongest in the evaluated commercial regime (0.819 AUROC, 0.910 AUARC) and ranks hard-answer model conflict with AUROC up to 0.982. The pooled prediction also improves accuracy over the average single model by 5.6 to 13.0 points. Over the full trajectory of a multi-step, multi-agent system, subagent collective uncertainty ranks system failures above chance (0.619 AUROC) and gives the best abstention ordering among the evaluated signals (0.699 AUARC).

Abstract

Hallucination Detection in VLMs

Explore similar work

CUSP: Decomposable Collective Uncertainty for Multi-Agent Multimodal Reasoning

UHP Detection: LVLMs have their Unique Hallucination Pattern in the Consistency Space

The Cost of Reasoning: Chain-of-Thought Induces Overconfidence in Vision-Language Models