Value of Information under Imprecise Probabilities: Decision-Rule-Specific Values and Fixed-Measure Envelopes on a Credal Set

Feed of new arXiv stat-ML papers covering theory and methods across uncertainty quantification, scalable sampling and generative modelling, manifold and graph methods, inverse problems, and detection of AI artifacts. Notable entries with security relevance: a unified detection framework using Mahalanobis distance scores for LLM-generated text, hallucination, watermark and adversarial-example detection (includes robust covariance estimators); a paper showing conformal prediction can under-cover minority classes in imbalanced, high-stakes domains (drug discovery) and proposing a class-conditonal