BEGIN:VCALENDAR
VERSION:2.0
PRODID:researchseminars.org
CALSCALE:GREGORIAN
X-WR-CALNAME:researchseminars.org
BEGIN:VEVENT
SUMMARY:Andrés Gómez (USC)
DTSTART:20200602T183000Z
DTEND:20200602T190000Z
DTSTAMP:20260423T021857Z
UID:DOTs/12
DESCRIPTION:Title: <a href="https://researchseminars.org/talk/DOTs/12/">Ou
 tlier detection in time series via mixed-integer conic quadratic optimizat
 ion</a>\nby Andrés Gómez (USC) as part of Discrete Optimization Talks\n\
 n\nAbstract\nWe consider the problem of estimating the true values of a Wi
 ener process given noisy observations corrupted by outliers. The problem c
 onsidered is closely related to the Trimmed Least Squares estimation probl
 em\, a robust estimation procedure well-studied from a statistical standpo
 int but poorly understood from an optimization perspective. In this paper 
 we show how to improve existing mixed-integer quadratic optimization formu
 lations for this problem. Specifically\, we convexify the existing formula
 tions via lifting\, deriving new mixed-integer conic quadratic reformulati
 ons. The proposed reformulations are stronger and substantially faster whe
 n used with current mixed-integer optimization solvers. In our experiments
 \, solution times are improved by at least two orders-of-magnitude.\n
LOCATION:https://researchseminars.org/talk/DOTs/12/
END:VEVENT
END:VCALENDAR