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:CMIH Hub seminar series
SUMMARY:Improving the Flexibility and Robustness of Deriva
 tive-Free Optimization Solvers - Lindon Roberts
DTSTART;TZID=Europe/London:20190308T140000
DTEND;TZID=Europe/London:20190308T150000
UID:TALK120973AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/120973
DESCRIPTION:Classical nonlinear optimisation algorithms requir
 e the availability of gradient evaluations for con
 structing local approximations to the objective an
 d testing for convergence. In settings where the o
 bjective is expensive to evaluate or noisy\, evalu
 ating the gradient may be too expensive or inaccur
 ate\, so cannot be used\; we must turn to optimisa
 tion methods which do not require gradient informa
 tion\, so-called derivative-free optimisation (DFO
 ). This has applications in areas such as climate 
 modelling\, hyperparameter tuning and generating a
 dversarial examples in deep learning. In this talk
 \, I will introduce DFO and discuss two software p
 ackages for DFO for nonlinear least-squares proble
 ms and general minimisation problems. I will descr
 ibe their novel features aimed at expensive and/or
  noisy problems\, and show their state-of-the-art 
 performance. Time permitting\, I will also show a 
 heuristic method which improves the ability of the
 se methods to escape local minima\, and show its f
 avourable performance on global optimisation probl
 ems.
LOCATION:MR11\,  Centre for Mathematical Sciences\, Wilberf
 orce Road\, Cambridge
CONTACT:J.W.Stevens
END:VEVENT
END:VCALENDAR