No-U-Turn Sampler

E898983

Hamiltonian Monte Carlo variant Markov chain Monte Carlo algorithm adaptive MCMC method

The No-U-Turn Sampler is an adaptive variant of Hamiltonian Monte Carlo that automatically tunes trajectory lengths to efficiently explore complex probability distributions without manual parameter selection.

Try in SPARQL Jump to: Statements Referenced by

Statements (49)

Predicate	Object
instanceOf	Hamiltonian Monte Carlo variant ⓘ Markov chain Monte Carlo algorithm ⓘ adaptive MCMC method ⓘ
abbreviation	NUTS NERFINISHED ⓘ
advantageOver	standard HMC with fixed trajectory length ⓘ
aimsTo	efficiently explore complex probability distributions ⓘ improve mixing of Markov chains ⓘ improve sampling efficiency ⓘ reduce random walk behavior in MCMC ⓘ
assumes	ability to compute gradients of log posterior ⓘ
basedOn	Hamiltonian Monte Carlo NERFINISHED ⓘ
category	gradient-based MCMC method ⓘ
comparedTo	standard Hamiltonian Monte Carlo ⓘ
field	Bayesian statistics ⓘ computational statistics ⓘ machine learning ⓘ
fullName	No-U-Turn Sampler NERFINISHED ⓘ
hasComponent	U-turn stopping rule ⓘ step size adaptation procedure ⓘ tree-building procedure ⓘ
hasFeature	adaptive trajectory length ⓘ automatic tuning of path length ⓘ dynamic integration time ⓘ no manual number of leapfrog steps selection ⓘ no manual trajectory length selection ⓘ stopping criterion based on U-turn detection ⓘ
hasVariant	NUTS with dual averaging step size adaptation NERFINISHED ⓘ
implementedIn	NumPyro NERFINISHED ⓘ PyMC NERFINISHED ⓘ Stan NERFINISHED ⓘ TensorFlow Probability NERFINISHED ⓘ Turing.jl NERFINISHED ⓘ
introducedBy	Andrew Gelman NERFINISHED ⓘ Matthew D. Hoffman NERFINISHED ⓘ
introducedInPublication	No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo NERFINISHED ⓘ
property	maintains detailed balance ⓘ produces asymptotically exact samples under regularity conditions ⓘ reduces need for manual tuning compared to basic HMC ⓘ suitable for high-dimensional parameter spaces ⓘ uses gradient information of log posterior ⓘ
publicationYear	2011 ⓘ
requires	differentiable log density ⓘ
targetDistribution	continuous probability distributions ⓘ
usedFor	Bayesian posterior inference ⓘ hierarchical Bayesian models ⓘ probabilistic modeling ⓘ sampling from posterior distributions ⓘ
uses	Hamiltonian dynamics ⓘ leapfrog integrator ⓘ

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

Hamiltonian Monte Carlo → generalization → No-U-Turn Sampler ⓘ