The research direction centers on derivations, simulator comparisons, and ablations on stability, identifiability, and reconstruction quality.
The central question is whether a learned dynamics surrogate can retain the structure needed for inverse problems while becoming faster and easier to optimize.
The visual language on the site treats this project as a field-line system: continuous dynamics, measured signals, and inferred latent states.