Machine Learning in Disordered Photonics

ended

The field of disordered photonics has increased its importance immensely over past decades as it finds widespread application in several fields from biomedical imaging, to solar energy harvesting, paint, pigments, food and cosmetic industry. However, the current development of highly scattering materials is often hindered by lack of ways to quantitatively predict and model their structural morphology and photonic properties. This action aims to characterize disordered photonic structures made of organic materials by analyzing their 3D structures using Gaussian Processes (GP) based machine learning techniques in conjunction with numerical optical simulations. The inherent randomness in the 3D arrangement of disordered photonics, makes them both intuitively and theoretically ideal to be modeled with GP. The novelty of this action consist of using state-of-the art GP method not only analyze 3D structures, but also to reconstruct them from lower dimensional data, like 2D images and spectroscopic data. Moreover by using the quantitative GP descriptors, we are able both generate input models for numerical simulations and using the feedback iteratively update those models to optimize them for high scattering. We expect that the complementary expertise in characterization and computational methods of the Host and the Researcher will produce not only invaluable insights, but also practical tools to characterize, quantify and exhaustively model and optimize complex photonic structures.