Do we understand what a neural network learns? More importantly, what does it mean to understand in this context. Despite unprecedented success in predictive performance and broad applicability across disciplines, this question remains largely unexplored.

We explore light-matter interactions by treating both light and matter quantum mechanically. In several regimes, such a description is essential and gives rise to phenomena that are fundamentally distinct from semiclassical methods. ​Our research leverages machine learning (ML) approaches for modeling the dynamics efficiently. Beyond predictive performance, we place particular emphasis on the interpretability of ML models, with a goal of understanding what neural networks learn and how their internal representations connect to the fundamental principles.