BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Robust Empirical Bayes for Gaussian Processes - Masha Naslidnyk (U
 niversity College London)
DTSTART:20230309T111500Z
DTEND:20230309T120000Z
UID:TALK196444@talks.cam.ac.uk
DESCRIPTION:For many modern statistical machine learning problems\, model 
 misspecification is pervasive and impactful. In particular\, inferences ar
 e not robust\, and uncertainty quantification becomes brittle. These issue
 s are exacerbated by nonparametric models like Gaussian Processes. While d
 istance-based estimation is a powerful remedy for this setting\, previousl
 y proposed distances between conditional distributions are intractable. To
  resolve this\, we introduce a computationally tractable distance on the s
 pace of conditional probability distributions we call expected maximum con
 ditional mean discrepancy. The theoretical properties of the resulting dis
 tance-based estimator are investigated in detail. While the estimator is o
 f general interest\, we focus on its application as a robust empirical Bay
 es estimator in Gaussian Process models. Specifically\, we demonstrate tha
 t it produces reliable uncertainty quantification for regression problems\
 , computer model emulation\, and Bayesian optimisation.&nbsp\;
LOCATION:Seminar Room 2\, Newton Institute
END:VEVENT
END:VCALENDAR