PhD: Efficient optimisation for fast simulation under budget constraints
Where: Leiden University, 
Netherlands
What:
Within the NWO AI4optics project, this PhD develops AI-driven optimisation methods for accelerating optical simulation, a key bottleneck in the design of imaging and illumination systems. AI4optics brings together experts in computer science, mathematics and physics to develop AI-based methods for solving inverse problems in high-tech optical systems. By making optical simulation faster and more efficient, the project contributes to improved optical design workflows across a wide range of high-tech applications.
A major challenge in ray-tracing simulation is selecting the most informative rays from very large candidate sets without sacrificing simulation accuracy. In this project, you will develop optimisation methods that efficiently explore these large search spaces by exploiting meaningful similarity measures between rays. Building on Leiden University's previous work on distance-based subset optimisation, originally developed for optimal optical filter selection, you will investigate structured sampling strategies together with quality-diversity and generative optimisation methods to accelerate simulation-driven optical design.
The project also explores learning-based optimisation to further reduce computational cost. You will investigate warm-starting optimisation from previously discovered solutions, transferring knowledge between related optimisation problems and removing redundant search across optimisation runs. In parallel, you will develop AI methods for efficiently predicting phase-space partitions and design novel heuristic algorithms for mixed-variable, multi-objective optimisation of imaging optics involving both continuous geometric parameters and discrete design choices such as lens materials.
Who: Anna Kononova, a.kononova@liacs.leidenuniv.nl
When: Until 2026-09-01 05:59
Presented at next GECCO?: yes