Improving the efficiency and robustness of black-box optimisation

Abstract

In classical nonlinear optimisation, the availability of first-order information is crucial to constructing accurate local models for the objective and finding descent directions. However, when the objective function is black-box and computationally expensive or stochastic, gradients may be too expensive to compute or too inaccurate to be useful. In this setting, derivative-free optimisation (DFO) provides an alternative class of algorithm. In this talk, I will present new techniques for model-based DFO, which yield an improvement in the efficiency and robustness of existing methods for general minimisation and nonlinear least-squares problems.

Date
26 Apr 2019
Event
2nd IMA and OR Society Conference on Mathematics of Operational Research
Location
Aston University
Avatar
Lindon Roberts
Lecturer

My research is in numerical analysis, particularly nonconvex and derivative-free optimization.