Black box optimization concerns the optimization of functions that can only be evaluated through numerical simulation and in which partial derivates are either not known or not defined. One way to optimize these functions consists of using consecutive function evaluations to build and refine a surrogate model, and then use this model to drive the optimization. Albeit very effective, this model-based approach is typically computationally expensive, becoming impractical when the number of input variables is greater than a few dozens and when multiple objectives are considered jointly.
This project, financed by WASP (Wallenberg AI, Autonomous Systems and Software Program), aims at introducing innovative algorithms and methodologies to overcome the limitations of black-box optimization. The research topic falls at the crossroads of the broader fields of black-box optimization and statistical machine learning. This project will develop statistical methods for modelling surrogate models. These models can then be queried using Bayesian optimization or similar techniques to identify values of process settings that optimize the quality of the multiple objectives, or that maximize the information gained from experiments. Using a Bayesian statistical approach augmented with user prior knowledge will enable combining information from different sources, e.g. experiments of different kinds, while accounting for the uncertainty involved in a rigorous and coherent manner. The new algorithms and methodologies will be tested on a variety of synthetic and real-world applications such as automated machine learning (AutoML), automated configuration of compilers and hardware design.
This project is part of a collaboration with Stanford University. The employee will be encouraged to apply to the Lund-Stanford exchange program to work closely with collaborators.