BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:809b8e924fcd65aa9a9a69a66caee2b7
DTSTAMP:20261019T185621Z
SUMMARY:Statistic Seminar: Martin Wainwright (MIT) – The Information Geom
 etry of Diffusion Sampling
DESCRIPTION:14:00 – 15:00 – Martin Wainwright (MIT)\nTitle:The Informat
 ion Geometry of Diffusion Sampling\nAbstract: Sampling from a high-dimensi
 onal distribution is a core challenge spanning statistics\, operations res
 earch\, applied mathematics and computer science. Diffusion-based methods 
 have led to dramatic advances: in generative AI\, they produce strikingly 
 realistic images\, video\, and text\, and they are finding growing use in 
 areas including biology\, drug design\, materials science\, and robotics. 
  In this talk\, we introduce a unified way of analyzing the performance o
 f both continuous diffusion samplers (based on Gaussian noise)\, and discr
 ete analogues. We introduce information-theoretic quantities that control 
 the Kullback–Leibler divergence of a given sampler.  They adapt natural
 ly to low dimensional structure in the underlying distribution\, and can b
 e substantially smaller than what ambient dimension alone would suggest. M
 oreover\, these path-based measures can be estimated from data\, providing
  a way to calibrate and improve the performance of diffusion samplers.\n 
 \n 
URL:https://www.imperial.ac.uk/events/215470/statistic-seminar-martin-wainw
 right-mit/
DTSTART;TZID=Europe/London:20261021T140000
DTEND;TZID=Europe/London:20261021T150000
LOCATION:HXLY144\, Huxley Building\, South Kensington Campus\, Imperial Col
 lege London\, London\, SW7 2AZ\, United Kingdom
END:VEVENT
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
DTSTART:20261021T140000
TZNAME:BST
TZOFFSETTO:+0100
TZOFFSETFROM:+0100
END:DAYLIGHT
END:VTIMEZONE
END:VCALENDAR