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:20211031T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20220327T020000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:calendar.23941.field_data.0@www.diag.uniroma1.it
DTSTAMP:20240619T075947Z
CREATED:20211212T170139Z
DESCRIPTION:Two-level stochastic optimization formulations have become inst
rumental in a number of machine learning contexts such as neural architect
ure search\, continual learning\, adversarial learning\, and hyperparamete
r tuning. Practical stochastic bilevel optimization problems become chall
enging in optimization or learning scenarios where the number of variables
is high or there are constraints.The goal of this work is twofold. First\
, we aim at promoting the use of bilevel optimization in large-scale learn
ing and we introduce a practical bilevel stochastic gradient method (BSG-1
) that requires neither lower level second-order derivatives nor system so
lves (and dismisses any matrix-vector products). Our BSG-1 method is close
to first-order principles\, which allows it to achieve a performance bett
er than those that are not\, such as DARTS. Second\, we develop bilevel st
ochastic gradient descent for bilevel problems with lower level constraint
s\, and we introduce a convergence theory that covers the unconstrained an
d constrained cases and abstracts as much as possible from the specifics o
f the bilevel gradient calculation. Joint work with Griffin Kent and Luis
Nunes Vicente. Preprint: https://arxiv.org/abs/2110.00604
DTSTART;TZID=Europe/Paris:20211221T110000
DTEND;TZID=Europe/Paris:20211221T110000
LAST-MODIFIED:20211212T181537Z
LOCATION:Aula Magna - DIAG
SUMMARY:Bilevel stochastic methods for optimization and machine learning: B
ilevel stochastic descent and DARTS - Tommaso Giovannelli (Lehigh Universi
ty\, Bethlehem\, PA\, USA)
URL;TYPE=URI:http://www.diag.uniroma1.it/node/23941
END:VEVENT
END:VCALENDAR