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:Isaac Newton Institute Seminar Series
SUMMARY:Multiscale methods and recursion in data science -
  Piotr Fryzlewicz (London School of Economics)
DTSTART;TZID=Europe/London:20180323T113000
DTEND;TZID=Europe/London:20180323T123000
UID:TALK102844AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/102844
DESCRIPTION:The talk starts on a general note: we first attemp
 t to define a "multiscale" method / algorithm as a
  recursive program acting on a dataset in a suitab
 le way. Wavelet transformations\, unbalanced wavel
 et transformations and binary segmentation are all
  examples of multiscale methods in this sense. Usi
 ng the example of binary segmentation\, we illustr
 ate the benefits of the recursive formulation of m
 ultiscale algorithms from the software implementat
 ion and theoretical tractability viewpoints. <br><
 br>We then turn more specific and study the canoni
 cal problem of a-posteriori detection of multiple 
 change-points in the mean of a piecewise-constant 
 signal observed with noise. Even in this simple se
 t-up\, many publicly available state-of-the-art me
 thods struggle for certain classes of signals. In 
 particular\, this misperformance is observed in me
 thods that work by minimising a "fit to the data p
 lus a penalty" criterion\, the reason being that i
 t is challenging to think of a penalty that works 
 well over a wide range of signal classes. To overc
 ome this issue\, we propose a new approach whereby
  methods learn from the data as they proceed\, and
 \, as a result\, operate differently for different
  signal classes. As an example of this approach\, 
 we revisit our earlier change-point detection algo
 rithm\, Wild Binary Segmentation\, and make it dat
 a-adaptive by equipping it with a recursive mechan
 ism for deciding "on the fly" how many sub-samples
  of the input data to draw\, and w here to draw th
 em. This is in contrast to the original Wild Binar
 y Segmentation\, which is not recursive. We show t
 hat this significantly improves the algorithm part
 icularly for signals with frequent change-points. 
 <br><br>Related Links<ul><li><a target="_blank" re
 l="nofollow" href="http://www-old.newton.ac.uk/cgi
 /https%3A%2F%2FCRAN.R-project.org%2Fpackage%3Dbrea
 kfast">https://CRAN.R-project.org/package=breakfas
 t</a> - R software package "breakfast" (provides a
 n implementation of Adaptive Wild Binary Segmentat
 ion)</li></ul>
LOCATION:Seminar Room 1\, Newton Institute
CONTACT:INI IT
END:VEVENT
END:VCALENDAR