Gradient-free Hamiltonian Monte Carlo with efficient Kernel exponential familiesReportar como inadecuado




Gradient-free Hamiltonian Monte Carlo with efficient Kernel exponential families - Descarga este documento en PDF. Documentación en PDF para descargar gratis. Disponible también para leer online.

Reference: Strathmann, H, Sejdinovic, D, Livingstone, S et al., (2015). Gradient-free Hamiltonian Monte Carlo with efficient Kernel exponential families.Citable link to this page:

 

Gradient-free Hamiltonian Monte Carlo with efficient Kernel exponential families

Abstract: We propose Kernel Hamiltonian Monte Carlo (KMC), a gradient-free adaptiveMCMC algorithm based on Hamiltonian Monte Carlo (HMC). On target densitieswhere classical HMC is not an option due to intractable gradients, KMC adaptivelylearns the target’s gradient structure by fitting an exponential family modelin a Reproducing Kernel Hilbert Space. Computational costs are reduced by twonovel efficient approximations to this gradient. While being asymptotically exact,KMC mimics HMC in terms of sampling efficiency, and offers substantial mixingimprovements over state-of-the-art gradient free samplers. We support our claimswith experimental studies on both toy and real-world applications, including ApproximateBayesian Computation and exact-approximate MCMC.

Peer Review status:Peer reviewedPublication status:PublishedVersion:Accepted ManuscriptConference Details: Advances in Neural Information Processing Systems

Bibliographic Details

Publisher: Massachusetts Institute of Technology Press

Publisher Website: http://mitpress.mit.edu/

Host: Advances in Neural Information Processing Systemssee more from them

Publication Website: http://papers.nips.cc/

Issue Date: 2015Identifiers

Urn: uuid:bc1fc6ae-4b39-408a-bf47-0a9c952ca8ba

Source identifier: 577243

Issn: 1049-5258 Item Description

Type: Conference;

Version: Accepted Manuscript Tiny URL: pubs:577243

Relationships





Autor: Strathmann, H - - - Sejdinovic, D - institutionUniversity of Oxford Oxford, MPLS, Statistics - - - Livingstone, S - - - Szabo, Z

Fuente: https://ora.ox.ac.uk/objects/uuid:bc1fc6ae-4b39-408a-bf47-0a9c952ca8ba



DESCARGAR PDF




Documentos relacionados