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:Non-parametric Bayesian Method and Maximum-A-Poste
riori Inference in Statistical Machine Translation
- Tsuyoshi Okita (Dublin City University)
DTSTART;TZID=Europe/London:20120502T111500
DTEND;TZID=Europe/London:20120502T121500
UID:TALK38018AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/38018
DESCRIPTION:Since recent sophisticated Machine Learning algori
thms implicitly handle various things\, practition
ers do not need to worry much about how to deploy
those algorithms in particular situations. However
\, if it comes to real-life data such as Statistic
al Machine Translation\, several things were worth
considering: 1) the underlying distribution may b
e better assumed to be the power-law distribution
rather than its i.i.d. counterpart\, 2) noise may
not be captured well as a simple Gaussian type (he
nce\, such noise assumption is not often embedded
in the ML algorithm)\, 3) available prior knowledg
e may not be sufficiently used\, and so forth. It
is noted that what kinds of non-Gaussian type nois
e we need to focus on and what kind of prior knowl
edge we need to target were not evident from the b
eginning (These issues would be quite difficult ev
en if we can exploit the domain experts. This is s
ince these require both the knowledge of the under
lying ML algorithm and the domain knowledge of the
area). We discuss two algorithms in the applicati
on area of Statistical Machine Translation: non-pa
rametric Bayesian method (hierarchical Pitman-Yor
process related topics) and Maximum-A-Posteriori i
nference. The first algorithm is related to the la
nguage model smoothing where 1) is concerned\, whi
le the second algorithm is related to the word ali
gnment where 2) and 3) are concerned.
LOCATION:Engineering Department\, CBL Room BE-438
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR