BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Machine Learning Reading Group @ CUED
SUMMARY:Hamiltonian Monte Carlo for Hierarchical Models -
Vidhi Lalchand
DTSTART;TZID=Europe/London:20191023T140000
DTEND;TZID=Europe/London:20191023T153000
UID:TALK133699AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/133699
DESCRIPTION:Hierarchical models provide a powerful framework f
or modelling and inference by defining second orde
r and third order probability distributions over p
arameters at different levels of the generative mo
del. Hamiltonian Monte Carlo (HMC) is one of the p
rimary tools for inference in hierarchical models.
While hierarchies provide modelling flexibility\,
they induce distinctive pathologies in the poster
ior that limit the efficiency of sampling algorith
ms like HMC. These pathologies can be best detecte
d by visualising the joint posterior\ngeometry thr
ough bivariate density plots and by HMC diagnostic
s. In this talk we will review HMC and its limitat
ions in the context of posterior inference in hier
archical models. We will discuss some common techn
iques to simplify the posterior geometry through\n
reparameterization that can significantly improve
sampling efficiency. We will also briefly review a
dvances like Riemann Manifold HMC that can address
some of the weaknesses of Euclidean HMC in sampli
ng from posterior geometries characterized by tigh
t correlations and\ndrastically changing curvature
.
LOCATION:Engineering Department\, CBL Room BE-438
CONTACT:Robert Pinsler
END:VEVENT
END:VCALENDAR