Small Business Information
Technology Service Corporation
1900 S. Sepulveda Blvd, Suite 300, Los Angeles, CA, 90025
AbstractThe discrimination of lethal RVs within the ballistic missile complex is a challenging task. In Phase I, TSC demonstrated a nonparametric Bayesian classifier that can efficiently estimate the Probability Density Function (PDF) used in the Bayes classifier. A novel method of mapping N-dimensional space to 1-dimensional space was developed to mitigate the "dimensionality curse". Feedback from the PDF estimate is employed to reduce the simulation and feature extraction processing load. In Phase I, TSC showed that our technique could estimate complex PDFs with good accuracy while substantially reducing the training requirements. TSC's classifier and training architecture is compatible with both continuous and statistically-dependent features and should outperform conventional Dynamic Bayes Networks (DBNs) in many cases. In Phase II, TSC will refine the PDF estimator, implement the full classifier training architecture, and compare discrimination performance to conventional DBN algorithms. TSC will also investigate any limitations of our mapping technique, analyze its accuracy and improve its speed. TSC will additionally seek measured missile flight test data for testing. In Phase III, TSC plans to transition this technology into the BMD Decision Architecture.
* information listed above is at the time of submission.