BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//prima-2022//speaker calendar//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VTIMEZOME
TZID:America/Vancouver
TZURL:http://tzurl.org/zoneinfo-outlook/America/Vancouver
X-LIC-LOCATION:America/Vancouver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP;TZID=America/Vancouver:20221209T103000
DTSTART;TZID=America/Vancouver:20221209T103000
DTEND;TZID=America/Vancouver:20221209T105500
UID:20221209T103000@prima2022.primamath.org
SUMMARY:From sparse polynomial approximation to optimal restart schemes for accelerating first-order optimization algorithms
DESCRIPTION:Sparse polynomial methods have, over the last decade, become widely used tools for approximating smooth, high-dimensional functions from limited data. Notable applications include parametric modelling and computational uncertainty quantification, discovering governing equations of dynamical systems, and high-dimensional PDEs. A standard sparse polynomial approximation scheme involves solving a convex optimization problem involving a $\ell^1$- or weighted $\ell^1$-norm. This is typically done using first-order methods. However, such methods generally converge very slowly, making them undesirable for the many applications that require high accuracy. Motivated by this problem, in this talk I will describe a general scheme for accelerating first-order optimization algorithms. This scheme applies to any convex optimization problem possessing a certain $\textit{approximate sharpness}$ condition and essentially any first-order optimization algorithm. For a wide range of problems, i.e., not just $\ell^1$-minimization-type problems, this scheme achieves either optimal rates of convergence or improves on previously established rates. To emphasize its usefulness and generality, I will showcase applications in compressive imaging and machine learning.
STATUS:CONFIRMED
LOCATION:Junior Ballroom C
END:VEVENT
END:VCALENDAR