“During my undergraduate studies in engineering physics, I was immediately hooked by the elegant and layered mathematical model for optics, ranging from straightforward paraxial optics to the more subtle photonics, and the wide variety of phenomena it could explain. I was eager to put my knowledge to work during internships in academia and industry where I did research on medical, panoramic and enhanced depth-of-field imaging, and developed a growing interest for research in optics.

“At the beginning of my graduate studies, with courses in both machine learning and lens design, I soon came to the realization that there was a lot to gain in lens design by casting the problem as a learning one, and to my surprise the literature held little information on this topic. Witnessing the strong emergence of deep learning, I asked myself: how can lens design benefit from this? As I found out, one answer to this question was to extrapolate from lens design databases, in the hope of extracting the features of what makes a lens design successful, and combine them to address other specifications or lens configurations.”

“Aside from implementing natural extensions to our framework (modeling aspherical surfaces, extending to other wavelengths, …), I want to create more data-driven tools for lens designers, such as the idea of “training” an optimizer on the landscape of typical lens design merit functions, or using reinforcement learning to add or remove lens elements as the optimization progresses.”

As of 2023, Geoffroi is a post-doctoral researcher in Artificial Intelligence and Optics at Princeton University.