This research is designed to integrate the following three areas: (1) Bayesian statistical inference and decision theory, (2) computational and numerical methods for statistical applications, and (3) applications of computational and statistical methodology to problems of modeling. At the core of this project is the adaptation of methods for the parallel statistics and decision theory, and to other numerical and computational methods. Two broad classes of problems to be studied that relate parallel processing to other computational problems are the use of a parallel system for the computation of high-dimensional integrals and the optimization of high- dimensional functions on a parallel system. In both classes of problems. the basic question is to determine an efficient way to decompose the computation across a collection of different processors operating in parallel. The methods, algorithms, and software to be developed for parallel processing will be useful in the determination of optimal or approximately optimal Bayesian solutions in a wide variety of sequential decision problems, including applications in robotics.
