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 @ CUED
SUMMARY:Bayesian Nonparametric Model for Power Disaggregat
ion - Isabel Valera (University Carlos III in Madr
id)
DTSTART;TZID=Europe/London:20140217T110000
DTEND;TZID=Europe/London:20140217T120000
UID:TALK50807AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/50807
DESCRIPTION:We propose an *infinite factorial unbounded-state
hidden Markov model* (HMM) through the constructio
n of a Bayesian nonparametric (BNP) prior over int
eger-valued matrices (in which each column represe
nts a Markov chain) with the property of presentin
g an infinite number of columns with an unbounded
number of states\, namely\, IFUHMM. First\, we ext
end the existent infinite factorial binary-state H
MM to allow for any number of states\, and derive
two Markov chain Monte Carlo (MCMC) and a variatio
nal inference algorithms. Then\, we modify this mo
del to allow an unbounded number of states and der
ive a new inference algorithm based on MCMC method
s that properly deals with the trade-off between t
he unbounded number of states and chains. Finally\
, we apply both the infinite factorial (nonbinary)
HMM and the IFUHMM to the *power disaggregation p
roblem* and show their ability to provide more int
erpretable representations of the data structure t
han the existing BNP approaches for HMMs.
LOCATION:Engineering Department\, CBL Room BE-438
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR