SBIR/STTR Award attributes
Timely and effective decision making to protect and employ US space-based resources in complex evolving situations depends on rapid and perceptive multi-source space situational awareness (SSA) in support of Battle Management Command and Control (BMC2). A robust machine learning (ML) SSA system needs to interface readily to new and different sources and types of evidence, to learn and recognize situational patterns from this constantly changing evidence across domains, and to robustly adapt, autonomously and with least cost and disruption, to the widest range of both anticipated and unanticipated changes in space phenomena. To address these requirements and provide a robust adaptive ML architecture, Charles River Analytics proposes to design an innovative Multi-source Relational Situational Intelligence (MRSI) system. MRSI constructs and maintains relational models of evolving multi-domain situations from dynamic multi-source evidence for future BMC2 SSA. MRSI uses probabilistic programming, based on our Figaro(TM) and Scruff(TM) probabilistic programming languages, to integrate domain knowledge with learning from data, enabling an SSA system to continually adapt and improve itself over time. The probabilistic models are organized in hierarchical predictive processing models that enable the SSA system to reason about high-level events from low-level sensor data.
