The best of both worlds: fast and accurate prediction of meta-optics with physics-guided machine learning
Viktor A. Podolskiy, Sean Lynch, Jacob LaMountain, Jie Bu, Bo Fan, Amogh Raju, Anuj Karpatne, Daniel M. Wasserman · 2024
We aim to address one of the fundamental limitations of machine learning (ML): its reliance on extensive training datasets by incorporating physics-based intuition and Maxwell-equation-based constraints into ML process. We show that physics-guided networks require significantly smaller datasets, enable learning outside the original training data, and provide improved prediction accuracy and physics consistency. The proposed approaches are illustrated on examples of photonic composites, from photonic crystals to hyperbolic metamaterials.