BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Scalable Sampling Using Annealed Algorithms - Saifuddin Syed (Univ
 ersity of Oxford)
DTSTART:20241120T110000Z
DTEND:20241120T123000Z
UID:TALK224758@talks.cam.ac.uk
CONTACT:Xianda Sun
DESCRIPTION:Generating samples from complex probability distributions is a
  fundamental challenge in statistical modelling and Bayesian statistics. I
 n practice\, this is generally impossible\, and we must introduce a simple
 r reference distribution\, such as a Gaussian\, and manipulate its density
  and samples to approximate the target. In general\, direct inference is r
 eliable when the reference is close to the target and fragile when it is n
 ot. Annealing is a popular technique motivated by this principle and intro
 duces a sequence of distributions that interpolates between the reference 
 and target\, ensuring the neighbouring distributions are close enough. An 
 annealing algorithm specifies how to traverse this bridge of distributions
  to incrementally transform samples from the reference into samples approx
 imating the target.\n\nIn this talk\, we will construct two computationall
 y dual annealing algorithms called Sequential Monte Carlo Samplers (SMC) a
 nd Parallel Tempering (PT)\, which propagate samples from the reference to
  the target using importance sampling and Metropolis-Hasting\, respectivel
 y. By analysing the variance of the normalising constant estimator\, we wi
 ll see how the performance scales with increasing runtime\, parallelism\, 
 memory\, and the difficulty of the inference problem. Notable\, we will id
 entify a critical phenomenon and explain why these algorithms are efficien
 t and can scale to tackle modern sampling problems. Finally\, we will prov
 ide a black-box algorithm to tune these algorithms efficiently and practic
 al guidelines for when to implement SMC versus PT.
LOCATION:Cambridge University Engineering Department\, CBL Seminar room BE
 4-38.
END:VEVENT
END:VCALENDAR