Koopman Generative Operator

A generative model for probabilistic time-series forecasting based on Koopman operator theory, performing linear dynamical evolution in a latent space. It combines flow matching to achieve efficient single-step generation and adaptively decomposes uncertainty.