latent dynamical models are commonly used to learn the distribution of a
latent dynamical process that represents a sequence of noisy data samples.
However, producing samples from such models with high fidelity is challenging
due to the complexity and variability of latent and observat