BEGIN:VCALENDAR
VERSION:2.0
PRODID:ILLC Website
X-WR-TIMEZONE:Europe/Amsterdam
BEGIN:VTIMEZONE
TZID:Europe/Amsterdam
X-LIC-LOCATION:Europe/Amsterdam
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700329T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701025T030000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:/NewsandEvents/Archives/2007/newsitem/1740/27-
 February-2007-HAI-Tea-Pieter-Adriaans
DTSTAMP:20070222T000000
SUMMARY:HAI-Tea, Pieter Adriaans
ATTENDEE;ROLE=Speaker:Pieter Adriaans
DTSTART;TZID=Europe/Amsterdam:20070227T150000
DTEND;TZID=Europe/Amsterdam:20070227T000000
LOCATION:Room C210, Roeterseiland, Nieuwe Achtergr
 acht 127, Amsterdam
DESCRIPTION:In this lecture I will present some re
 cent work I did with Paul Vitanyi and Ceriel Jacob
 s on the application of the MDL (Minimum Descripti
 on Length) principle to grammar induction. We have
  studied MDL in terms of two-part code optimizatio
 n and randomness deficiency. These notions will be
  explained in the lecture. In this framework we sh
 owed that: 1) Shorter code not necessarily leads t
 o better theories, e.g. the randomness deficiency 
 does not decrease monotonically with the MDL code,
  2) contrary to what is suggested by the results o
 f Gold:1967 there is no fundamental difference bet
 ween positive and negative data from an MDL perspe
 ctive, 3) MDL is extremely sensitive to the correc
 t calculation of code length. Using these ideas we
  have implemented a MDL variant of the EDSM algori
 thm. The results show that although MDL works well
  as a global optimization criterion, it falls shor
 t of the performance of algorithms that evaluate l
 ocal features of the problem space. MDL can be des
 cribed as a global strategy for featureless learni
 ng.   For more information, see http://homepages.c
 wi.nl/~paulv/papers/perils.pdf and http://staff.sc
 ience.uva.nl/~pietera/ALS/background/lncs_icgi-mdl
 .pdf.  For more information on HAI-Tea lectures, s
 ee http://www.science.uva.nl/onderwijs/studieprogr
 amma/haitea/.
X-ALT-DESC;FMTTYPE=text/html:\n      <p>\n        
 In this lecture I will present some recent work I 
 did with\n        Paul Vitanyi and Ceriel Jacobs o
 n the application of the MDL\n        (Minimum Des
 cription Length) principle to grammar\n        ind
 uction. We have studied MDL in terms of two-part c
 ode\n        optimization and randomness deficienc
 y. These notions will be\n        explained in the
  lecture. In this framework we showed that: 1)\n  
       Shorter code not necessarily leads to better
  theories,\n        e.g. the randomness deficiency
  does not decrease monotonically\n        with the
  MDL code, 2) contrary to what is suggested by the
 \n        results of Gold:1967 there is no fundame
 ntal difference\n        between positive and nega
 tive data from an MDL perspective, 3)\n        MDL
  is extremely sensitive to the correct calculation
  of code\n        length.  Using these ideas we ha
 ve implemented a MDL variant\n        of the EDSM 
 algorithm. The results show that although MDL\n   
      works well as a global optimization criterion
 , it falls short\n        of the performance of al
 gorithms that evaluate local features\n        of 
 the problem space. MDL can be described as a globa
 l\n        strategy for featureless learning.\n   
    </p>\n    \n      <p>For more information, see\
 n        <a target="_blank" href="http://homepages
 .cwi.nl/~paulv/papers/perils.pdf">http://homepages
 .cwi.nl/~paulv/papers/perils.pdf</a> and\n        
 <a target="_blank" href="http://staff.science.uva.
 nl/~pietera/ALS/background/lncs_icgi-mdl.pdf">http
 ://staff.science.uva.nl/~pietera/ALS/background/ln
 cs_icgi-mdl.pdf</a>.<br />\n        For more infor
 mation on HAI-Tea lectures, see\n        <a target
 ="_blank" href="http://www.science.uva.nl/onderwij
 s/studieprogramma/haitea/">http://www.science.uva.
 nl/onderwijs/studieprogramma/haitea/</a>.\n      <
 /p>\n    
URL:/NewsandEvents/Archives/2007/newsitem/1740/27-
 February-2007-HAI-Tea-Pieter-Adriaans
END:VEVENT
END:VCALENDAR