BEGIN:VCALENDAR
PRODID:-//Columba Systems Ltd//NONSGML CPNG/SpringViewer/ICal Output/3.3-
 M3//EN
VERSION:2.0
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
DTSTAMP:20260115T121645Z
DTSTART:20260121T140000Z
DTEND:20260121T150000Z
SUMMARY:SQUIDS Seminar: Sampling as Bandits: Evaluation-Efficient Design 
 for Black-Box Densities
UID:{http://www.columbasystems.com/customers/uom/gpp/eventid/}d1fc-mkfez6
 le-l2p15
DESCRIPTION:Speaker: Takuo Matsubara (University of Edinburgh)\n\nAbstrac
 t: We introduce Bandit Importance Sampling (BIS)\, a novel class of impo
 rtance sampling methods designed for scenarios where the target density 
 is computationally expensive to evaluate. In contrast to adaptive import
 ance sampling\, which optimises a proposal distribution\, BIS directly o
 ptimises samples through a sequential selection process combined with mu
 lti-armed bandits. BIS serves as a general framework accommodating bespo
 ke bandit strategies. Crucially\, we establish guarantees of weak conver
 gence for the weighted samples\, which hold regardless of the choice of 
 bandit strategy. We present a practical strategy that leverages Gaussian
  process surrogates to guide sample selection\, enabling efficient explo
 ration of the parameter space with a minimal number of target evaluation
 s. BIS yields accurate approximations with fewer target evaluations\, de
 monstrating superior performance across multimodal\, heavy-tailed distri
 butions\, and real-world Bayesian inference tasks involving computationa
 lly intensive models.
STATUS:TENTATIVE
TRANSP:TRANSPARENT
CLASS:PUBLIC
LOCATION:Frank Adams Room 2\, Alan Turing Building\, Manchester
END:VEVENT
END:VCALENDAR