BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Regularization for Optimal Transport and Dynamic Time Warping Dist
ances
DTSTART;VALUE=DATE-TIME:20180412T123000Z
DTEND;VALUE=DATE-TIME:20180412T133000Z
DTSTAMP;VALUE=DATE-TIME:20200222T092743Z
UID:indico-event-2617@indico.math.cnrs.fr
DESCRIPTION:Machine learning deals with mathematical objects that have str
ucture. Two common structures arising in applications are point clouds / h
istograms\, as well as time series. Early progress in optimization (linear
and dynamic programming) have provided powerful families of distances bet
ween these structures\, namely Wasserstein distances and dynamic time warp
ing scores. Because they rely both on the minimization of a linear functio
nal over a (discrete) space of alignments and a continuous set of coupling
s respectively\, both result in non-differentiable quantities. We show how
two distinct smoothing strategies result in quantities that are better be
haved and more suitable for machine learning applications\, with applicati
ons to the computation of Fréchet means.\n\nhttps://indico.math.cnrs.fr/e
vent/2617/
LOCATION:UMPA
URL:https://indico.math.cnrs.fr/event/2617/
END:VEVENT
END:VCALENDAR