Strathmann, H;
Sejdinovic, D;
Livingston, S;
Schuster, I;
Lomeli Garcia, M;
Szabo, Z;
Andrieu, C;
(2016)
Kernel techniques for adaptive Monte Carlo methods.
Presented at: Greek Stochastics Workshop on Big Data and Big Models, Tinos, Greek.
Preview |
Text
strathmann16kernel_slides.pdf Download (1MB) | Preview |
Abstract
We introduce a general kernel-informed Monte Carlo algorithm family for fast sampling from Bayesian posterior distributions. Our focus is on the highly challenging Big Models regime, where posteriors often exhibit strong nonlinear correlations and the evaluation of the target density (and its gradient) is either analytically intractable or computationally expensive. To construct efficient sampling schemes in these cases, application of adaptive MCMC methods learning the target geometry becomes critical. We present how kernel methods can be embedded into the adaptive MCMC paradigm enabling to construct rich classes of proposals with attractive convergence and mixing properties. Our ideas are exemplified for three popular sampling techniques: Metropolis-Hastings, Hamiltonian Monte Carlo and Sequential Monte Carlo.
| Type: | Conference item (Presentation) |
|---|---|
| Title: | Kernel techniques for adaptive Monte Carlo methods |
| Event: | Greek Stochastics Workshop on Big Data and Big Models |
| Location: | Tinos, Greek |
| Dates: | 10 - 13 July 2016 |
| Open access status: | An open access version is available from UCL Discovery |
| Publisher version: | http://www.stochastics.gr/meetings/theta/ |
| Language: | English |
| UCL classification: | UCL UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Life Sciences UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Life Sciences > Gatsby Computational Neurosci Unit |
| URI: | https://discovery.ucl.ac.uk/id/eprint/1496927 |
Archive Staff Only
 |
View Item |
