BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:David Widmann (Uppsala University) DTSTART:20221124T141500Z DTEND:20221124T150000Z DTSTAMP:20260422T155025Z UID:gbgstats/7 DESCRIPTION:Title: Calibration of probabilistic predictive models\nby David Widmann (Upp sala University) as part of Gothenburg statistics seminar\n\nLecture held in MVL14.\n\nAbstract\nMost supervised machine learning tasks are subject to irreducible prediction errors. Probabilistic predictive models address this limitation by providing probability distributions that represent a be lief over plausible targets\, rather than point estimates. Such models can be a valuable tool in decision-making under uncertainty\, provided that t he model output is meaningful and interpretable. Calibrated models guarant ee that the probabilistic predictions are neither over- nor under-confiden t. In the machine learning literature\, different measures and statistical tests have been proposed and studied for evaluating the calibration of cl assification models. For regression problems\, however\, research has been focused on a weaker condition of calibration based on predicted quantiles for real-valued targets. In this paper\, we propose the first framework t hat unifies calibration evaluation and tests for general probabilistic pre dictive models. It applies to any such model\, including classification an d regression models of arbitrary dimension. Furthermore\, the framework ge neralizes existing measures and provides a more intuitive reformulation of a recently proposed framework for calibration in multi-class classificati on. In particular\, we reformulate and generalize the kernel calibration e rror\, its estimators\, and hypothesis tests using scalar-valued kernels\, and evaluate the calibration of real-valued regression problems.\n LOCATION:https://researchseminars.org/talk/gbgstats/7/ END:VEVENT END:VCALENDAR