SCoOP: Semantic Consistent Opinion Pooling for Uncertainty Quantification in Multiple Vision-Language Model Systems
This addresses the reliability issue in multimodal AI systems by enabling effective uncertainty-aware aggregation, though it is incremental as it builds on existing opinion pooling methods for a specific multi-VLM context.
The paper tackles the problem of uncertainty amplification and hallucination risk when combining multiple Vision-Language Models (VLMs) by proposing SCoOP, a training-free uncertainty quantification framework that achieves an AUROC of 0.866 for hallucination detection and an AURAC of 0.907 for abstention, outperforming baselines by 10-13% and 7-9% respectively.
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.