Northwest Fisheries Science Center

Display All Information

Document Type: Journal Article
Center: NWFSC
Document ID: 8770
Title: Locally adaptive nonparametric smoothing with Markov random fields and shrinkage priors
Author: James R. Faulkner, Vladimir N. Minin
Publication Year: 2017
Journal: Bayesian Analysis
Volume: Advance publication
Keywords: Bayesian, horseshoe prior, lasso, penalized regression, Hamiltonian Monte Carlo, Levy process,

We present a locally adaptive nonparametric curve fitting method that operates within a fully Bayesian framework.  This method uses shrinkage priors to induce sparsity in order-k differences in the latent trend function, providing a combination of local adaptation and global control.  Using a scale mixture of normals representation of shrinkage priors, we make explicit connections between our method and kth order Gaussian Markov random field smoothing.  We call the resulting processes shrinkage prior Markov random fields (SPMRFs).  We use Hamiltonian Monte Carlo to approximate the posterior distribution of model parameters because this method provides superior performance in the presence of the high dimensionality and strong parameter correlations exhibited by our models.  We compare the performance of three prior formulations using simulated data and find the horseshoe prior provides the best compromise between bias and precision.  We apply SPMRF models to two benchmark data examples frequently used to test nonparametric methods.  We find that this method is flexible enough to accommodate a variety of data generating models and offers the adaptive properties and computational tractability to make it a useful addition to the Bayesian nonparametric toolbox.

URL1: The next link will exit from NWFSC web site
Theme: Recovery and rebuilding of marine and coastal species
Foci: Develop methods to use physiological, biological and behavioral information to predict population-level processes.