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

2026-03-24

AI Summary

This work addresses the heightened hallucination risk in ensembles of vision-language models (VLMs), which often stems from amplified uncertainty. To mitigate this, the authors propose SCoOP—a training-free uncertainty quantification framework that explicitly models collective uncertainty at the multi-VLM system level for the first time. SCoOP achieves reliable aggregation through semantic-consistent uncertainty-weighted linear opinion pooling, enabling both hallucination detection and automatic abstention on high-uncertainty samples, thereby overcoming the limitations of single-model uncertainty estimation. Evaluated on ScienceQA, SCoOP significantly outperforms baseline methods in hallucination detection (AUROC = 0.866) and abstention performance (AURAC = 0.907), while introducing only microsecond-level inference overhead.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Large Vision Models
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

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 multi-VLM systems through uncertainty-weighted linear opinion pooling. 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.

Problem

• uncertainty quantification
• vision-language models
• hallucination detection
• model aggregation
• multimodal reasoning

Innovation

• Uncertainty Quantification
• Vision-Language Models
• Opinion Pooling
• Hallucination Detection
• Multimodal Reasoning