Sampling from the Horseshoe
When
Where
Speaker: Andrew Arnold, Program in Applied Mathematics, University of Arizona
Abstract: The so-called "horseshoe prior" [*] has achieved some traction in the Bayesian statistics community due to its design, which promotes sparsity among posterior variables while preserving values of significant posterior variables. I will present on the general theory of Hamiltonian Monte Carlo and importance sampling, then move on to the application of these techniques to a high-dimensional horseshoe. Those interested in application of Bayesian MCMC are encouraged to attend and discuss their problems (and the inevitable woes of MCMC!). The talk will also be of interest to the sparsity-seeking sort.
[*] Carvalho, Polson, Scott. Handling Sparsity via the Horseshoe. 2009.