P22-073 NeuroPlex++

To achieve overmatch in C4I versus peer adversaries, Army intelligence analysts need support in detecting and classifying instances of adversary tactics, techniques, and procedures (TTP) execution across domains. Complex Event Processing (CEP) tools need improvement to meet this requirement, in order to detect entity behaviors and movements that may be observed by different sensors, or separated in time by minutes or longer. TTPs also change frequently, and must be recognized when they occur in different operational environments and across domains. An additional challenge is that relatively little training data may exist to help a system learn to detect and recognize instances of TTP execution. The proposed NeuroPlex++ tool builds on previous work and uses a hybrid approach to solve this problem, leveraging deep neural networks for mapping unstructured high dimensional sensor data into symbolic percepts embedded in space and time with associated uncertainty, along with a logical layer to enable the direct incorporation of human domain knowledge to enable CEP and TTP recognition across operational environments and domains, meeting the retraining objective of this topic.