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:20161030T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20160327T020000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:calendar.7338.field_data.0@www.corsodrupal.uniroma1.it
DTSTAMP:20230605T201045Z
CREATED:20160627T221414Z
DESCRIPTION:Trust-region methods are a broad class of methods for continuou
soptimization that finds application in a variety of problemsand contexts.
The basic principle consists of iteratively optimizinga model of the obje
ctive function in a restricted region. In particular\,they have been studi
ed and applied for problems without using derivatives\,where models are bu
ilt solely by sampled function values.Trust-region derivative-free methods
are guaranteed toconverge for deterministic smooth functions\, in the sen
seof generating a sequence or run of iterates converging to criticality(of
first and second order type). Such a convergence is calledglobal as there
is no assumption on the starting point.It has also been proved that the o
rder of complexity\, or theglobal rate in which criticality decays\, match
es thederivative-based case.In the deterministic non-smooth case\, and giv
en some knowledgeof the non-smooth structure\, it is possible to design gl
oballyconvergent approaches with appropriated complexity\, either bysmooth
ing the original function in some parametrized way or bymoving a compositi
ve non-smooth structure directly to thetrust-region subproblem.Trust-regio
n methods can be based as well on probabilistic modelsof the objective fun
ction\, thus considering the derivative-freecase where sample points are r
andomly generated. Such methodsexhibit similar properties of convergence a
nd complexity as thoseusing deterministic models\, now not for all runs bu
t for aa set of those that occurs with probability one.Finally\, we are ob
serving now the first attempts for thestochastic case\, where the objectiv
e function can only beobserved\, and possibly approximated by sample avera
ging.This talk will attempt to overview all such developments andpoint out
what still remains unknown.
DTSTART;TZID=Europe/Paris:20160628T143000
DTEND;TZID=Europe/Paris:20160628T143000
LAST-MODIFIED:20200521T211813Z
LOCATION:Aula A4 - DIAG Via Ariosto
SUMMARY:SEMINARIO: Recent Progress on Derivative-Free Trust-Region Methods
- Prof. Luis Nunes Vicente\, University of Coimbra
URL;TYPE=URI:http://www.corsodrupal.uniroma1.it/node/7338
END:VEVENT
END:VCALENDAR