Computational feasibility of tailoring the proposed RMHMC to more complex GP models

Determine whether the proposed RMHMC implementation with a soft‑absolute Hessian metric and dynamic eigendecomposition maintains reasonable computation time when tailored to hierarchical Gaussian‑process models augmented with reversible‑jump mechanisms and to time‑series models combined with sequential Monte Carlo.

Background

The authors discuss extending their RMHMC implementation to more elaborate real‑world settings, including sparsity‑inducing reversible‑jump mechanisms for Bayesian multiple‑kernel models and structured time‑series models that would require coupling RMHMC with sequential Monte Carlo.

They explicitly state that it is currently unknown whether such tailored versions will remain computationally feasible, highlighting a key practical question for broader applicability.