BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Artificial Intelligence Research Group Talks (Comp
uter Laboratory)
SUMMARY:Non-convex Optimisation Using the Polyak-Łojasiewi
cz Inequality - Edoardo Calvello (Imperial College
)
DTSTART;TZID=Europe/London:20200204T130000
DTEND;TZID=Europe/London:20200204T140000
UID:TALK135394AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/135394
DESCRIPTION:We consider the idea of solving a non-convex optim
isation problem by adding a large enough strongly-
convex function to make the objective function con
vex. This allows the use of simpler convex optimis
ers\, yet\, given the large strong-convexity const
ant required\, yields a value closer to the minimu
m of the function added than that of the objective
one. We try to fix this with the novel idea of in
stead adding a function that makes the objective s
atisfy the Polyak-Łojasiewicz (PL) inequality\, a
much weaker condition than strong-convexity. Build
ing on previous work\, we construct an optimisatio
n algorithm relying on this method. We find that a
much smaller multiplicative constant is needed fo
r convergence to a minimum. We attempt to find and
prove convergence rates and computational complex
ity and test which algorithm yields a more accurat
e minimum.
LOCATION:LT2\, Computer Laboratory\, William Gates Building
CONTACT:Mateja Jamnik
END:VEVENT
END:VCALENDAR