BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:1
UID:UW-Physics-Event-8135
DTSTART:20230208T170000Z
DTEND:20230208T181500Z
DTSTAMP:20231003T162822Z
LAST-MODIFIED:20230206T210257Z
LOCATION:Online Seminar: Please sign up for our mailing list at www.ph
ysicsmeetsml.org for zoom link
SUMMARY:Bayesian Renormalization: An explicit correspondence between s
tatistical inference and exact renormalization\, Physics ∩ ML Semina
r\, Marc Klinger\, University of Illinois Urbana-Champaign
DESCRIPTION:Renormalization is a ubiquitous tool in theoretical physic
s used to understand the role of scale in organizing natural phenomena
. In this presentation we will report on a new information theoretic p
erspective for understanding the Exact Renormalization Group (ERG) thr
ough the intermediary of Bayesian Statistical Inference. This connecti
on is facilitated by the Dynamical Bayesian Inference scheme\, which e
ncodes Bayesian inference in the form of a one parameter family of pro
bability distributions solving an integro-differential equation derive
d from Bayes’ law. Utilizing the picture of an ERG flow as a functio
nal diffusion process\, we arrive at a dictionary outlining how renorm
alization can be understood as an inverse process relative to a Dynami
cal Bayesian inference scheme. A particularly salient feature of this
correspondence is that it identifies the role of Fisher geometry in pr
oviding an emergent scale for “Bayesian Renormalization” that is r
elated to the precision with which nearby points in model space can be
differentiated. We comment on the usefulness of this identification i
n data science applications including possible implementations of “B
ayesian Renormalization” as a tool for refining diffusion learning t
echniques.
URL:https://www.physics.wisc.edu/events/?id=8135
END:VEVENT
END:VCALENDAR