BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/842
DTSTAMP:20230914T125940Z
SUMMARY:Beyond Parametric Models: Robust Machine Learning with Permutation-
based Models
DESCRIPTION:Speaker: Ashwin Pananjady (University of California\, Berkeley
\nDepartment of Electrical Engineering \nand Computer Science \nBerkeley\,
CA \nUnited States of America)\n\nAbstract: \nParametric models offer sim
plicity and interpretability\, and have been instrumental in driving progr
ess in machine learning over the last few decades. In this talk\, I will d
escribe progress on supplementing these models by explicitly incorporating
underlying permutations in two canonical machine learning settings -- ran
king and regression -- and show that the resulting "permutation-based" mod
els are significantly richer but still preserve the advantages of parametr
ic models. Focussing first on the ranking aspect\, I will show the utility
of permutation-based models in estimating the results of pairwise compari
sons and aggregating survey responses\, both of which are standard crowdso
urcing tasks. In particular\, I will present algorithms that are significa
ntly more robust than their parametric counterparts for ranking from parti
al pairwise comparisons\, while also being rate-optimal in some settings.
In addition\, I will describe our recent progress on characterizing the st
atistical and computational limits of this problem -- which are conjecture
d to differ -- by presenting the first algorithm that makes progress on cl
osing a conjectured statistical-computational gap. I will also briefly tou
ch upon the regression aspect\, showing that such a permutation-based appr
oach is suitable for modelling correspondence tasks in computer vision\, a
nd allows us to design rate-optimal estimators for this classical problem.
\n\nBio: Ashwin Pananjady is a fourth year Ph.D. student in the Department
of Electrical Engineering and Computer Sciences at the University of Cali
fornia\, Berkeley\, advised by Martin Wainwright and Thomas Courtade. His
interests are in machine learning\, optimization\, information theory\, an
d statistics. He obtained his B.Tech. in Electrical Engineering from IIT M
adras in 2014\, and graduated with the Governor's Gold Medal.\n
URL:https://www.tcs.tifr.res.in/web/events/842
DTSTART;TZID=Asia/Kolkata:20180108T140000
DTEND;TZID=Asia/Kolkata:20180108T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
END:VCALENDAR