BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Date iCal//NONSGML kigkonsult.se iCalcreator 2.20.2//
METHOD:PUBLISH
X-WR-CALNAME;VALUE=TEXT:Eventi DIAG
BEGIN:VTIMEZONE
TZID:Europe/Paris
BEGIN:STANDARD
DTSTART:20181028T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
RDATE:20191027T030000
TZNAME:CET
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20190331T020000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:calendar.18169.field_data.0@www.diag.uniroma1.it
DTSTAMP:20240524T001321Z
CREATED:20190505T102647Z
DESCRIPTION:Mathematical Programming solvers operate several complex algori
thmic components\, which a user can control by means of parameters. Findin
g a good solver parameter configuration is often necessary to achieve a go
od solution (or at least a feasible one) for a given instance. However\, d
ue to the high number of parameters\, configuring a solver is usually nont
rivial.We propose a methodology\, based on machine learning and optimizati
on\, for selecting a solver configuration for a given instance. First\, we
employ a set of solved instances and configurations in order to learn a p
erformance function of the solver (Performance Map Learning Phase -- PMLP)
. Secondly\, we solve a mixed-integer nonlinear program in order to find t
he best algorithmic configuration based on the performance function (Confi
guration Space Search Problem -- CSSP). The main novelty of this work is t
hat the mathematical program\, that we optimize to configure the solver\,
embeds an explicit formulation of the mathematical properties of the chose
n machine learning predictor. The approach outlined was tested and evaluat
ed on a set of instances of the Hydro Unit Commitment problem\, solved usi
ng the general-purpose IBM ILOG CPLEX solver. We used the Support Vector R
egression technique for the PMLP and the Bonmin solver to optimize the CSS
P
DTSTART;TZID=Europe/Paris:20190517T150000
DTEND;TZID=Europe/Paris:20190517T160000
LAST-MODIFIED:20200521T211813Z
LOCATION:DIAG - Via Ariosto 25
SUMMARY:MORE@DIAG: Algorithmic Configuration By Learning And Optimization -
Gabriele Iommazzo
URL;TYPE=URI:https://www.diag.uniroma1.it/node/18169
END:VEVENT
END:VCALENDAR