aiqbench analysis
This research exposes a critical evaluation artifact in Evidential Deep Learning where Out-of-Distribution detection metrics are artificially inflated by uncontrolled differences in class cardinality between training and test sets. The study demonstrates that standard AUROC and AUPR scores can vary significantly without any actual change in model predictions, necessitating stricter definitions of In-Distribution and OOD contexts for LLMs. For practitioners relying on uncertainty quantification, this highlights the urgent need to normalize class counts before comparing detection performance across different datasets.
Rethinking Vacuity for OOD Detection in Evidential Deep Learning
arXiv:2605.06382v2 Announce Type: replace Abstract: Vacuity, or Uncertainty Mass (UM), is commonly used as a metric to evaluate Out-of-Distribution (OOD) detection in Evidential Deep Learning (EDL). It generally involves dividing the number of classes ($K$) by the total strength of belief ($S$) of the model's predictions, where $S$ is derived from summing the Dirichlet parameters. As such, UM is sensitive to the cardinality of $K$. As a result, when comparing In Distribution (ID) and OOD results, it is important that $K_{\mathrm{ID}}$ and $K_{\mathrm{OOD}}$ are equal; something that is not always ensured in practice. We provide an empirical demonstration of how results for AUROC and AUPR can substantially differ when class cardinality between ID and OOD differs by 1, with AUROC differing by as much as 0.346 and AUPR by 0.634 for standard EDL, and AUROC by 0.427 and AUPR by 0.745 for IB-EDL (both from Implementation B, Llama3-8B, ARC-E). Our findings isolate an evaluation artefact: when $K$ differs between ID and OOD, AUROC/AUPR can be artificially inflated without any change in model predictions. We further discuss the evaluation of EDL over causal language models using Multiple-Choice Question-Answer (MCQA) datasets and argue for clearer definitions of ID and OOD in this context. Our primary contribution is an empirical and theoretical demonstration that vacuity-based OOD detection in EDL-fine-tuned LLMs is highly sensitive to uncontrolled differences in evaluated class cardinality.
Abstract:Vacuity, or Uncertainty Mass (UM), is commonly used as a metric to evaluate Out-of-Distribution (OOD) detection in Evidential Deep Learning (EDL). It generally involves dividing the number of classes ($K$) by the total strength of belief ($S$) of the model's predictions, where $S$ is derived from summing the Dirichlet parameters. As such, UM is sensitive to the cardinality of $K$. As a result, when comparing In Distribution (ID) and OOD results, it is important that $K_{\mathrm{ID}}$ and $K_{\mathrm{OOD}}$ are equal; something that is not always ensured in practice. We provide an empirical demonstration of how results for AUROC and AUPR can substantially differ when class cardinality between ID and OOD differs by 1, with AUROC differing by as much as 0.346 and AUPR by 0.634 for standard EDL, and AUROC by 0.427 and AUPR by 0.745 for IB-EDL (both from Implementation B, Llama3-8B, ARC-E). Our findings isolate an evaluation artefact: when $K$ differs between ID and OOD, AUROC/AUPR can be artificially inflated without any change in model predictions. We further discuss the evaluation of EDL over causal language models using Multiple-Choice Question-Answer (MCQA) datasets and argue for clearer definitions of ID and OOD in this context. Our primary contribution is an empirical and theoretical demonstration that vacuity-based OOD detection in EDL-fine-tuned LLMs is highly sensitive to uncontrolled differences in evaluated class cardinality.
Artificial Intelligence (cs.AI)
arXiv:2605.06382 [cs.AI]
(or arXiv:2605.06382v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2605.06382 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 314 words
Related coverage
Rethinking Epistemic Uncertainty in Node Classification through Information Growth
arXiv · Oct 6, 2026
Variational Inference for Evidential Deep Learning
arXiv · Jul 20, 2026
Robust Adversarial Quantification via Conflict-Aware Evidential Deep Learning
arXiv · Oct 6, 2026
Robust Evidential Learning Through Latent Consistency
arXiv · Oct 2, 2026