Efficient Global Optimization in Multiobjective Problems

The ParEGO algorithm and similar approaches use Gaussian Process surrogate models to do multiobjective optimization efficiently in the number of function evaluations. We would like to improve the scalability of these approaches so that both more objectives and more decision variables can be readily tackled. We would also like to improve convergence properties by understanding more about how much the model should be trusted in later evaluations. This project will exploit and extend cutting edge research from the machine learning community to achieve these aims.

PhD candidates with an interest in machine learning and optimization should apply. A strong mathematical background is desirable for this project.