Giannis Daras on posterior sampling

- Giannis Daras posted about posterior sampling methods that refine approximate distributions and add bias penalties when the sampler diverges from the true posterior. - His thread covered discrete sampling in latent spaces and neural data‑to‑energy Schrödinger bridges, claiming theoretical contraction guarantees toward the true distribution. - The posts frame sampling as a practical route to better Bayesian posteriors for molecular and generative tasks. (x.com 1) (x.com 2)