Skip to main content Skip to navigation

Calendar


Show all calendar items

Hector McKimm: Regeneration-enriched Hamiltonian Dynamics

- Export as iCalendar
Hamiltonian Monte Carlo (HMC) is a version of the Metropolis-Hastings algorithm, used for Markov chain Monte Carlo, which proposes new states by approximately simulating Hamiltonian dynamics. Restore processes are a class of continuous-time Markov processes, defined by enriching some underlying process with regenerations. The Hamiltonian-Restore process is a Restore process, created by introducing regenerations into a jump process that moves according to approximate Hamiltonian dynamics. The process is non-reversible: instead of relying on a Metropolis-Hastings accept-reject step, regenerations ensure the process converges to the correct target distribution.

Show all calendar items