BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VTIMEZONE
TZID:Europe/Vienna
BEGIN:DAYLIGHT
DTSTART:20250330T030000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20251026T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260424T143509Z
UID:688cb5b4ed9ea669729057@ist.ac.at
DTSTART:20250915T110000
DTEND:20250915T120000
DESCRIPTION:Speaker: Pragya Sur\nhosted by Marco Mondelli\nAbstract: Min-no
 rm interpolators naturally emerge as implicit regularized limits of modern
  machine learning algorithms. Recently\, their out-of-distribution risk wa
 s studied when test samples are unavailable during training. However\, in 
 many applications\, a limited amount of test data is typically available d
 uring training. Properties of min-norm interpolation in this setting are n
 ot well understood. In this talk\, I will present a characterization of th
 e risk of pooled min-L2-norm interpolation under covariate and concept shi
 fts. I will show that the pooled interpolator captures both early fusion a
 nd a form of intermediate fusion. Our results have several implications. F
 or example\, under concept shift\, adding data always hurts prediction whe
 n the signal-to-noise ratio is low. However\, for higher signal-to-noise r
 atios\, transfer learning helps as long as the shift-to-signal ratio lies 
 below a threshold that I will define. Our results also show that under cov
 ariate shift\, if the source sample size is small relative to the dimensio
 n\, heterogeneity between domains improves the risk. Time permitting\, I w
 ill introduce a novel anisotropic local law that allows to achieve some of
  these characterizations and is of independent interest in random matrix t
 heory. This is based on joint work with Yanke Song and Sohom Bhattacharya.
LOCATION:Office Bldg West / Ground floor / Heinzel Seminar Room (I21.EG.101
 )\, ISTA
ORGANIZER:swiddman@ist.ac.at
SUMMARY:Pragya Sur: Data Integration: Challenges and Opportunities for Inte
 rpolation Learning under Distribution Shifts
URL:https://talks-calendar.ista.ac.at/events/6006
END:VEVENT
END:VCALENDAR