sparse-identification

FLARE is a forced latent autoencoder that discovers compact response coordinates and sparse input-dependent latent dynamics from high-dimensional observations of forced physical systems, enabling long-horizon forecasting under unseen inputs.