Emulating quantum dynamics with neural networks via knowledge distillation

We introduce an efficient training framework for constructing machine learning-based emulators and demonstrate its capability by training an artificial neural network to predict the time evolution of quantum wave packets propagating through a potential landscape. This approach is based on the idea of knowledge distillation and uses elements of curriculum learning. It works by constructing a set of simple, but rich-in-physics training examples (a curriculum). These examples are used by the emulator to learn the general rules describing the time evolution of a quantum system (knowledge distillation). We show that this emulator is capable of learning the rules of quantum dynamics from a curriculum of simple training examples (wave packet interacting with a single rectangular potential barrier), and subsequently generalizes this knowledge to solve more challenging cases (propagation through an arbitrarily complex potential landscape). Furthermore, we demonstrate, that using this framework we can not only make high-fidelity predictions, but we can also learn new facts about the underlying physical system, detect symmetries, and measure relative importance of the contributing physical processes.