SEMANTIC GRAPHS FOR PATTERN DETECTION IN SENSOR DATA FROM MONITORING REGIMES

New monitoring regimes for future arms control treaties may generate large quantities of heterogeneous, noisy data that will be difficult for the unassisted analyst to use in verifying compliance. Traditional databases struggle to capture the multifaceted relationships expected in monitoring regime data. Instead, patterns of sensor outputs with complex relationships can be more intuitively expressed using semantic graph queries that better capture the sensor activity. These patterns can be defined using combinations of subject matter expertise, probabilistic techniques, and machine learning algorithms. The result is a hybrid analytical approach that accounts for both expert opinion of how sensors should operate in ideal scenarios as well as the noise and variability of humans interacting with sensors in operational environments. In this paper, we describe our motivation for using semantic graphs as the foundation for a hypothetical arms control monitoring regime, our approach for defining semantic graph models and parameters, and a simplified case study using the Sandia Arms Control Testbed. We conclude with lessons learned from our initial case studies, and propose areas of future research including potential applications to international safeguards, physical protection, and border security.