BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:CCM Colloq: Adrian Seyboldt (PyMC-Labs)\, on Fisher HMC sampling
DTSTART:20250219T150000Z
DTEND:20250219T160000Z
DTSTAMP:20260607T110900Z
UID:indico-event-4037@indico.flatironinstitute.org
DESCRIPTION:Presenter: Adrian Seyboldt (PyMC-Labs)\nTitle: If Only My Post
 erior Were Normal: Introducing Fisher HMC\nAbstract: Markov Chain Monte Ca
 rlo (MCMC) methods struggle with complex posteriordistributions. This talk
  introduces Fisher HMC\, a method that adaptivelytransforms parameter spac
 es to improve sampling efficiency of Hamiltonian MCMC.We use the Fisher di
 vergence to reshape the posterior\, so that we can explore itmore effectiv
 ely. I'll discuss the theoretical foundations\, and show how thisframework
  can be used to find better mass matrices\, or how we can usecomputational
 ly intensive normalizing flows for more adaptive transformations. Iwill sh
 ow how this is implemented in nutpie\, and can be used to improve sampling
 of PyMC and Stan models.\nBio: Adrian Seyboldt is a core PyMC developer an
 d statistics consultant withPyMC-Labs. He works on developing probabilisti
 c programming methods\, withexpertise in Bayesian computational techniques
  and scientific softwaredevelopment.\n\nhttps://indico.flatironinstitute.o
 rg/event/4037/
LOCATION:3rd floor classrm (Flatiron)
URL:https://indico.flatironinstitute.org/event/4037/
END:VEVENT
END:VCALENDAR