Formal cost separation and posterior fidelity of latent random-field sampling
Determine whether separating posterior conditioning from output-grid evaluation in a fixed-dimensional latent representation of random fields can be specified precisely enough to predict a measurable computational cost, and determine how faithfully the resulting latent posterior represents the true posterior when evaluated against exact references rather than other approximations.
References
For random fields, two questions therefore remain open: whether the separation of conditioning from the output grid can be specified precisely enough to predict a measurable cost, and how faithful the resulting posterior is when assessed against exact references rather than against other approximations.
— Latent Annealed Posterior Sampling with a Variance-Preserving Bridge for Sparsely Observed Random Fields
(2610.20659 - Lei et al., 14 Oct 2026) in Section 1, Introduction