Robust Classification through Deep Learning and Dynamic Multi-Entity Bayesian Reasoning

Missile defense faces the challenges of rapidly maturing and evolving complex threats, possessing capabilities which require the use of all available resources to successfully detect, track and identify the lethal objects. Future performance will rely on multiple sensors such as ground and sea based radars and electro-optical and infrared sensors for target recognition. It is crucial to develop a multi-sensor framework capable of processing the wealth of information these sophisticated systems can provide while also accounting for missing, noisy, or corrupted data. ExoAnalytic Solutions has partnered with Oregon State University to develop an advanced multi-sensor, multi-information -driven classifier for robust threat identification. We propose to do this through a unique combination of innovative advances in deep, hierarchical machine learning together with recurrent Deep Learning Neural Network (DLNN) methods for sensor fusion and Dynamic Multi-Entity Bayesian Networks (MEBNs) for whole-scene and contextual reasoning. Approved for Public Release 16-MDA-8620 (1 April 16)