CCM Colloquium: Nawaf Bou-Rabee (Rutgers University)

Date & Time

Title: Adjustable Randomized Time Integrators for Hamiltonian Monte Carlo


Abstract: Hamiltonian Monte Carlo is a gradient-based Metropolis-Hastings algorithm with a proposal distribution that involves ceil(T/h) steps of a time integrator of Hamiltonian dynamics; where T and h are the duration and time-step-size hyperparameters.  Here we show how to incorporate time integrator randomization into the proposal distribution and present L^2 Wasserstein mixing time guarantees for the corresponding unadjusted algorithm.  These guarantees illustrate the potential role of time integrator randomization in HMC.

Advancing Research in Basic Science and MathematicsSubscribe to Flatiron Institute announcements and other foundation updates