Online Program Home
My Program

Abstract Details

Activity Number: 524
Type: Topic Contributed
Date/Time: Wednesday, August 3, 2016 : 10:30 AM to 12:20 PM
Sponsor: Survey Research Methods Section
Abstract #319787 View Presentation
Title: Particle Swarm Optimization--Assisted Metropolis Hastings Algorithms
Author(s): Matthew Simpson* and Christopher Wikle and Scott H. Holan
Companies: University of Missouri and University of Missouri and University of Missouri
Keywords: Bayesian estimation ; Markov chain Monte Carlo ; Official statistics ; Optimization ; Spatial data
Abstract:

Fitting high-dimensional dependent data models is a challenging endeavor and typically requires some form of dimension reduction or stylized estimation algorithm. Particle swarm optimization (PSO) refers to a class of heuristic optimization algorithms that exploit analogies with animal flocking behavior in order obtain optima without strong conditions on the objective function. Bare bones PSO (BBPSO) is a particularly simple PSO algorithm that depends only the particle's personal best location, the particle's group best location, and Gaussian noise where "best'" is defined in terms of the objective function. We introduce a class of PSO algorithms based on BBPSO, termed adaptively tuned BBPSO (AT-BBPSO), by adding a scale parameter to the algorithm and which we dynamically tune using an analogy with random walk Metropolis. Our proposed algorithm is embedded within a Metropolis Hastings algorithm to provide an efficient proposal distribution and thus improves mixing and convergence. In order to illustrate this optimization method and compare it to alternatives, we provide a simulation study and apply it to a high-dimensional spatial problem arising from official statistics.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2016 program

 
 
Copyright © American Statistical Association