Bayesian Evolutionary Analysis with Sparse Training

To address the Missile Defense Agency (MDA) need for cognitive synthesis of current and emerging sensor data sources, Physical Optics Corporation (POC) has developed and demonstrated the feasibility of a new Bayesian Evolutionary Analysis with Sparse Training (BEAST) software system in Phase I. The Phase I BEAST system design combines outputs of Aegis classifier, radar measurements, and intelligence data to improve target classification and tracking capabilities, with low reliance on prior measurements. Semantics-based knowledge representation was developed to bridge the gap between abstract intelligence data and field measurements for target recognition. A Bayesian inference engine and truthing process was also developed to evolve threat behavior hypotheses based on the missile defense sensor measurements and post-flight analysis. These Phase I results demonstrated BEASTs feasibility to enhance Aegiss target recognition capabilities, set the groundwork for Phase II development, and defined a pathway towards future integration and testing of BEAST. In Phase II, we plan to enhance BEAST software, mature its algorithms and inference process, and integrate them into a Phase II prototype for testing, evaluation, and demonstration. A transition plan will be developed with the prime contractor to integrate and test BEAST in a testbed in Phase III. Approved for Public Release | 16-MDA-8951 (15 December 16)