BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//University of Liverpool Computer Science Seminar System//v2//EN
BEGIN:VEVENT
DTSTAMP:20260922T122037Z
UID:Seminar-dept-1259@lxserverM.csc.liv.ac.uk
ORGANIZER:CN=Lutz Oettershagen:MAILTO:Lutz.Oettershagen@liverpool.ac.uk
DTSTART:20250304T130000
DTEND:20250304T140000
SUMMARY:School Seminar Series
DESCRIPTION:Alessandro Varsi: The Quest for Optimal Parallel Monte Carlo Methods\n\nMonte Carlo (MC) methods are fundamental in Bayesian inference and Machine Learning, enabling sampling from complex posterior distributions. Markov Chain Monte Carlo (MCMC), first introduced in 1953, has been the dominant approach but suffers from being inherently sequential. In 1993, Sequential Monte Carlo (SMC) emerged as a parallelizable alternative, yet the resampling step (and its redistribution component) remained difficult to scale efficiently.\n\n\n\nFor years, parallel resampling was believed to be infeasible. However, breakthroughs in Distributed Memory (DM) architectures led to Optimal Parallel Redistribution (OPR) for fixed-size samples, culminating in an optimal O(logN) algorithm in 2021. Yet, recent evidence reveals that OPR for variable-size samples is NP-complete in DM, posing a major challenge for scalable Bayesian inference. In Shared Memory (SM), OPR for variable-size samples remains achievable, but an efficient DM solution is an open problem. \n\n\n\nThis talk explores past breakthroughs, fundamental limitations, and current challenges of parallel Monte Carlo methods.\n\n\n\nThis talk will be streamed on teams: https://teams.microsoft.com/l/meetup-join/19%3ameeting_ZjEwZmJiMGEtZTk2MC00OWExLTgwYzAtMTNkMGM0YWYwNDE0%40thread.v2/0?context=%7b%22Tid%22%3a%2253255131-b129-4010-86e1-474bfd7e8076%22%2c%22Oid%22%3a%22ea14c976-751c-48ac-9b37-112f41cfb44b%22%7d\n\n\n\nMeeting ID: 382 773 593 185\n\nPasscode: tX2QF9pn\n\nhttps://www.csc.liv.ac.uk/research/seminars/abstract.php?id=1259
LOCATION:ELEC204, 2th Floor Lecture Theatre EEE
END:VEVENT
END:VCALENDAR