analiza aiqbench
This research introduces a geometric approach to hallucination detection in LLMs and VLMs by modeling the response manifold as a density ridge derived from hidden state trajectories. The method achieves a significant 5-20 point gain in AUROC over existing unsupervised and supervised probes while maintaining robustness when calibration labels are severely limited. For practitioners facing data scarcity, this technique offers a promising alternative to standard entropy-based or probe-based detectors that typically degrade under such constraints.
Density Ridge Selective Prediction for LLM and VLM Hallucination Detection under Calibration Label Scarcity
arXiv:2606.10198v1 Announce Type: new Abstract: Hallucination detection in large language and vision-language models is increasingly framed as selective prediction, where a detector assigns a confidence score and abstains when confidence is low. Unsupervised sampling detectors (Semantic Entropy, EigenScore) avoid labels but plateau in quality, while supervised probes (SAPLMA) attain stronger in-distribution scores yet degrade sharply when calibration labels are scarce. We recover the response manifold of an LLM as the density ridge of a kernel density estimate built on a six-dimensional kinematic feature map of hidden state generation trajectories. A test generation is scored by the negated Euclidean distance from its projected feature point to the nearest ridge vertex, yielding a low-dimensional geometric skeleton of the stochastic output distribution. We evaluate against Semantic Entropy, SAR, EigenScore, SAPLMA, and log-probability on seven QA benchmarks (HaluEval-QA, TriviaQA, GSM8K, POPE, ScienceQA, A-OKVQA) using nine text and vision LLMs in a deliberately label-scarce protocol ($n_{\text{cal}}{=}200$ queries, $N{=}5$ generations). Our ridge-based score beats on AUROC with 5-20 points gain, while demonstrating tempered degradation under calibration-label scarcity.
Abstract:Hallucination detection in large language and vision-language models is increasingly framed as selective prediction, where a detector assigns a confidence score and abstains when confidence is low. Unsupervised sampling detectors (Semantic Entropy, EigenScore) avoid labels but plateau in quality, while supervised probes (SAPLMA) attain stronger in-distribution scores yet degrade sharply when calibration labels are scarce. We recover the response manifold of an LLM as the density ridge of a kernel density estimate built on a six-dimensional kinematic feature map of hidden state generation trajectories. A test generation is scored by the negated Euclidean distance from its projected feature point to the nearest ridge vertex, yielding a low-dimensional geometric skeleton of the stochastic output distribution. We evaluate against Semantic Entropy, SAR, EigenScore, SAPLMA, and log-probability on seven QA benchmarks (HaluEval-QA, TriviaQA, GSM8K, POPE, ScienceQA, A-OKVQA) using nine text and vision LLMs in a deliberately label-scarce protocol ($n_{\text{cal}}{=}200$ queries, $N{=}5$ generations). Our ridge-based score beats on AUROC with 5-20 points gain, while demonstrating tempered degradation under calibration-label scarcity.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2606.10198 [cs.LG]
(or arXiv:2606.10198v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2606.10198 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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.
Text integral sub licența deschisă a publicației · 269 cuvinte
Subiecte conexe
Overconfidence and Calibration in Medical VQA: Empirical Findings and Hallucination-Aware Mitigation
arXiv · 1 oct. 2026
Do LLM Recommenders Know When They're Hallucinating? Auditing Confidence Calibration in Catalog Faithfulness
arXiv · 25 aug. 2026
Credal Large Language Models for Semantic Commitment under Uncertainty
arXiv · 25 aug. 2026
Density Ridge Selective Prediction for LLM and VLM Hallucination Detection under Calibration Label Scarcity
arXiv · 12 iun. 2026