Dynamically Persistent Remote Inference of Nuclear
Facility Activity: Challenges and Approach

Monitoring and characterization of nuclear facilities is an essential activity of nuclear nonproliferation, materials control, and safeguards. Such inferences are best supported by plentiful, persistent, close-range sensors operating under control of the inference system, but those resources are not always available for real-world nonproliferation problems. We identify challenges and associated mitigation strategies related to integrating few, non-persistent, remote, third-party sensors into an autonomous monitoring and inference system. We present promising results from applying a prototype ML/AI system employing the proposed strategies to two testbed facilities and discuss ongoing efforts to improve knowledge management and update, uncertainty quantification, and ML/AI model interpretation and explanation.