Concave-Convex PDMP-based sampling

Recently nonreversible samplers based on simulating piecewise deterministic Markov processes (PDMPs) have shown potential for efficient sampling in Bayesian inference problems. However, there remains a lack of guidance how to best implement these algorithms. If implemented poorly, the computational costs event times can outweigh statistical efficiency dynamics. Drawing adaptive rejection literature, we propose concave-convex thinning approach process, which call CC-PDMP. This provides general guide constructing bounds that may be used facilitate PDMP-based sampling. A key advantage this method is its additive structure—adding decompositions yields decomposition. makes construction modular, as given decomposition class likelihoods and family priors, they combined construct posterior. We show our simple leads computationally thinning. Our well suited local PDMP simulation where conditional independence target exploited potentially huge gains. provide an R package compare with existing approaches events literature. Supplementary materials article are available online.