BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:22.10.2021 Sebastan Schmon (Maths): Simulation-based inference wit
h path signatures
DTSTART;VALUE=DATE-TIME:20211022T150000Z
DTEND;VALUE=DATE-TIME:20211022T170000Z
DTSTAMP;VALUE=DATE-TIME:20211204T142700Z
UID:indico-event-1056@conference.ippp.dur.ac.uk
DESCRIPTION:Scientific models described by computer simulators often lack
a tractable likelihood function\, precluding the use of standard likelihoo
d-based statistical inference. \n\nA plethora of statistical and machine
learning approaches have been developed to tackle this problem\, such as a
pproximate Bayesian computation and likelihood/posterior approximations. W
hat those methods have in common is the aim to connect real world data wit
h parameters of the underlying simulator.\n\nHowever\, effective measures
that can link simulated and observed data are generally difficult to const
ruct\, particularly for time series data which are often high-dimensional
and structurally complex. In this talk\, we discuss the use of path signat
ures as a natural candidate feature set for constructing summary statistic
s and distances between time series data for use in simulation-based infer
ence algorithms\, for example in approximate Bayesian computation. Our exp
eriments show that such an approach can generate more accurate approximate
Bayesian posteriors than existing techniques for time series models.\n\nz
oom link: https://durhamuniversity.zoom.us/j/93730370636?pwd=T2V4ekkvai95
K2paSlNqV21IMXRWUT09\n\n \nhttps://conference.ippp.dur.ac.uk/event/1056/
URL:https://conference.ippp.dur.ac.uk/event/1056/
END:VEVENT
END:VCALENDAR