Worked examples

Each worked example reproduces a full workflow from the diff-soup paper at a size that fits one workstation GPU. The configuration cell of each notebook lists the paper's settings next to the ones it runs.

• Learning constitutive laws from a phase-separation movie: the chemical potential and diffusivity of a phase-separating binary mixture, inferred with their uncertainty from one noisy movie of the Cahn-Hilliard dynamics.
• Learning a catalyst activity field: how the spatially varying catalyst activity inside a three-dimensional packed-bed reactor, learned from noisy concentrations by a neural network trained through the reactor simulation.