BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Centre for Mathematical Sciences - ECPv5.4.0.2//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Centre for Mathematical Sciences
X-ORIGINAL-URL:http://math-sciences.org
X-WR-CALDESC:Events for Centre for Mathematical Sciences
BEGIN:VTIMEZONE
TZID:UTC
BEGIN:STANDARD
TZOFFSETFROM:+0000
TZOFFSETTO:+0000
TZNAME:UTC
DTSTART:20230101T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=UTC:20230215T160000
DTEND;TZID=UTC:20230215T170000
DTSTAMP:20230401T143159
CREATED:20220921T083738Z
LAST-MODIFIED:20230210T141741Z
UID:4277-1676476800-1676480400@math-sciences.org
SUMMARY:Michael Faulkner (Bristol) - Statistical physics and its sampling algorithms: how it works and why a statistician should care
DESCRIPTION:Sampling algorithms are commonplace in statistics and machine learning – in particular\, in Bayesian computation – and have been used for decades to enable inference\, prediction and model comparison in many different settings. They are also widely used in statistical physics\, where many popular sampling algorithms first originated [1\, 2]. At a high level\, the goals within each discipline are the same – to sample from and approximate statistical expectations with respect to some probability distribution – but the motivations\, nomenclature and methods of explanation differ significantly. This has led to challenges in communicating between the fields\, and indeed the fundamental goals of one field are often misunderstood in the other. In this talk\, we elucidate statistical physics for the statistician\, with a particular emphasis on phase transitions in order to demonstrate that physicists tend to study probability models as functions of thermodynamic hyperparameters such as the temperature. \nWe then move on to sampling algorithms in general\, with a particular focus on the Metropolis [1] and molecular-dynamics [2] algorithms. Indeed\, certain key aspects of statistical physics have led to innovations in sampling algorithms that inform the Bayesian world. In particular\, one could argue that the Swendsen-Wang [3]\, Wolff [4] and event-chain Monte Carlo [5\, 6] algorithms were all developed in response to the onset of nonergodicity (with respect to both physical and Metropolis dynamics) at certain phase transitions. The final part of this talk therefore focusses on ergodicity breaking with respect to the Metropolis algorithm\, and how these alternative ergodic sampling algorithms were developed in response to this phenomenon. The aim here is to demonstrate that this key aspect of statistical physics has informed a fundamental goal of Bayesian computation — the development of efficient\, multi-purpose and ergodic sampling algorithms. \n \n[1] Metropolis\, Rosenbluth\, Rosenbluth\, Teller and Teller\, J. Chem. Phys. 21 1087 (1953) \n[2] Alder and Wainwright\, J. Chem. Phys. 27 1208 (1957) \n[3] Swendsen and Wang\, Phys. Rev. Lett. 58 86 (1987) \n[4] Wolff\, Phys. Rev. Lett. 62 361 (1989) \n[5] Bernard\, Krauth and Wilson\, Phys. Rev. E 80 056704 (2009) \n[6] Michel\, Mayer and Krauth\, EPL (Europhys. Lett.) 112 20003 (2015)
URL:http://math-sciences.org/event/michael-faulkner-bristol-tbc/
LOCATION:Room 101\, 2-5 Kirkby Place\, Plymouth\, PL4 6DT\, United Kingdom
CATEGORIES:Applied Mathematics,Statistics and Data Science,Theoretical Physics
END:VEVENT
END:VCALENDAR