BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Improving the Flexibility and Robustness of Derivative-Free Optimi
 zation Solvers - Lindon Roberts
DTSTART:20190308T140000Z
DTEND:20190308T150000Z
UID:TALK120973@talks.cam.ac.uk
CONTACT:J.W.Stevens
DESCRIPTION:Classical nonlinear optimisation algorithms require the availa
 bility of gradient evaluations for constructing local approximations to th
 e objective and testing for convergence. In settings where the objective i
 s expensive to evaluate or noisy\, evaluating the gradient may be too expe
 nsive or inaccurate\, so cannot be used\; we must turn to optimisation met
 hods which do not require gradient information\, so-called derivative-free
  optimisation (DFO). This has applications in areas such as climate modell
 ing\, hyperparameter tuning and generating adversarial examples in deep le
 arning. In this talk\, I will introduce DFO and discuss two software packa
 ges for DFO for nonlinear least-squares problems and general minimisation 
 problems. I will describe their novel features aimed at expensive and/or n
 oisy problems\, and show their state-of-the-art performance. Time permitti
 ng\, I will also show a heuristic method which improves the ability of the
 se methods to escape local minima\, and show its favourable performance on
  global optimisation problems.
LOCATION:MR11\,  Centre for Mathematical Sciences\, Wilberforce Road\, Cam
 bridge
END:VEVENT
END:VCALENDAR