Physics-informed machine learning
We will discuss several modern physics-informed approaches to data-driven dynamical systems and architecture design. The key principles that will guide our exploration will be parsimony and symmetry. Based on those principles we will discover correct dynamical descriptions of complicated systems only from (sparse and noisy) data and modern deep-learning architectures only from several basic ingredients: symmetry, stability and scale separation. Presented methods are an integral part of modern data-driven physics discovery and state-of-the-art deep-learning system design. We will provide Jupyter notebooks (or Colab) with a demonstration code on simple examples.
