Concise, accurate descriptions of physical systems through their conserved
quantities abound in the natural sciences. In data science, however, current
research often focuses on regression problems, without routinely incorporating
additional assumptions about the system that generated the data. Here, we
propose to explore a particular type of underlying stru