
Bayesian Logistic Regression Using Sequential Posterior Simulation 307
14.6 Conclusion
Sequential posterio r simulation algorithms complement earlier approaches to
the simulation approximation of posterior moments. Graphics processing units
have become a convenient and cost- e ffective platform for scientific computing,
potentially acceler ating computing speed by orders of magnitude for parallel
algorithms. One of the appealing fea tures of SPS is the fact that it is well
suited to GPU computing. The work reported here uses an SPS a lg orithm
designed to exploit that potential.
The multinomial logistic regression model, the focus of this chapter, is