Scalable Inference Algorithms for the BMD Decision Architecture

The Missile Defense Agency's (MDA) Hercules program is developing a Ballistic Missile Defense (BMD) Decision Architecture (DA). The DA serves as a technology platform in which to model, analyze, simulate and experiment to assess the capability envelopes of alternative missile defense system of systems configurations and operational protocols. In the DA, Bayesian Networks (BNs) are used to simultaneously represent many alternative time series of hypotheses about the state of a dynamic missile defense situation. BNs are computational objects consisting of a graphical data structure of random variables, and a solution algorithm that can compute the values of the joint probability distribution of the random variables in the BN based on dynamically observed evidence, such as sensor reports.

For this effort, IET is developing a suite of hybrid, exact+approximate Bayesian network solution algorithms that can robustly and dynamically trade-off accuracy against time and memory requirements of the solution algorithm. These algorithms will enable scaling of the Missile Defense Agency's Decision Architecture to simulate and experiment with the expected solution complexity of fully modeled Decision Architecture Bayesian networks. .