Discrimination

The performance of the Ballistic Missile Defense System depends on a number of factors; one of the most important is the ability to accurately and efficiently discriminate between threats (reentry vehicles) and non-threats (decoys). The DECISIVE ANALYTICS Corporation (DAC) team proposes to develop a discrimination system based on a novel machine learning algorithm called Maximum Variance Unfolding (MVU). The MVU algorithm combines the attributes of data compression and feature extraction, effectively learns nonlinear transformations such as changes in object morphology and dynamic object motion, and can be cast in the form of a semi-definite program which guarantees an optimal solution to the problem. We will illustrate the proficiency of our approach by learning high fidelity, high resolution radar signatures of a dynamic target model and testing the discrimination system on scenarios that confound the current Discrimination Fusion Engine (DFE). Phase II of this project will focus on integrating our discrimination system with our novel dynamic, hybrid Bayesian network model for combining tracking and discrimination.