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:Statistics
SUMMARY:The aggregation problems in learning theory - Guil
laume Lecue\, CNRS\, Universite Paris-Est Marne-l
a-vallee
DTSTART;TZID=Europe/London:20111118T160000
DTEND;TZID=Europe/London:20111118T170000
UID:TALK32518AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/32518
DESCRIPTION:Given a finite class F of functions there are thre
e aggregation problems:\n1) the problem of Model S
election aggregation: construct a procedure having
\na risk as close as possible to the best element
in F\,\n2) the problem of Convex aggregation: cons
truct a procedure having a risk as\nclose as possi
ble to the best element in the convex hull of F\,\
n3) the problem of Linear aggregation: construct a
procedure having a risk as\nclose as possible to
the best element in the linear span of F.\n\nWe wi
ll prove that empirical risk minimization is optim
al for the Convex and\nLinear aggregation problems
but sub-optimal for the Model Selection\naggregat
ion problem. Then we will construct an optimal agg
regation procedure\nfor the Model Selection aggreg
ation.\n
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0W
B
CONTACT:Richard Samworth
END:VEVENT
END:VCALENDAR