Full Posterior Inference

Background: Model limitations resulting from the chosen inference method in a time series prediction framework are points for future methodological advancement.

Question / Future Work: The MAP objective currently identifies a single, high-probability future trajectory. Future work should aim to characterize the full posterior distribution p(𝒚|𝒙) through methods like MCMC sampling in the latent space or using diffusion-based posterior approximations to provide a more complete uncertainty description.