Connecting the Dots: Numerical Randomized Hamiltonian Monte Carlo with State-Dependent Event Rates

Numerical generalized randomized Hamiltonian Monte Carlo is introduced, as a robust, easy to use and computationally fast alternative conventional Markov chain methods for continuous target distributions. A wide class of piecewise deterministic processes generalizing Randomized HMC (Bou-Rabee Sanz-Serna) by allowing state-dependent event rates defined. Under very mild restrictions, such will have the desired distribution an invariant distribution. Second, numerical implementation processes, based on adaptive integration second order ordinary differential equations (ODEs) considered. The yields approximate, yet highly robust algorithm that, unlike Carlo, enables exploitation complete trajectories (hence, title). proposed may yield large speedups improvements in stability relative relevant benchmarks, while incurring biases that are negligible overall errors. Granted access high-quality ODE code, methodology both implement use, even challenging high-dimensional Supplementary materials this article available online.