Development Of Novel Approaches To Anomaly Detection And Surety For Safeguards Data - Year Two And Three Results

The first phase of the Novel Approaches to Anomaly Detection and Surety for Safeguards Data project which considers the applicability for international safeguards of three core data analysis and management methods was presented at the INMM Annual Meeting in 2020. Year One of the project saw three major accomplishments. The first accomplishment was the prioritization and selection of anomaly detection methods to improve and extend the existing Grammar Compression (GC) method. One of the key results has been the development of a new method that combines GC with ensemble learning to perform robust and efficient anomaly detection in time series data. The second accomplishment was the down-selection of technologies and data for the prototype Distributed Ledger Technology (DLT) system. We have introduced and described a framework by which adoption tradeoffs of DLT for improved Continuity of Knowledge are being objectively evaluated. And the third accomplishment was an assessment of the viability of Multi-Party Computing (MPC) via a study of test scenarios to evaluate how easily anomalies in raw data sequences convert through a garbled circuit. These three approaches are natural complements to one another, as DLT and GC-based anomaly detection can be used on traditional safeguards data sources or other available data, and the MPC component allows for exploration of nontraditional data sources in a manner that will protect sensitive operator information. This paper outlines the work performed in Year Two and Three of the project and highlights results on: (1) the development of software tools implementing selected anomaly detection methods to extend and improve the existing GC method, (2) the development and evaluation of a software tool implementing the first version of the prototype DLT system, and (3) the application of the MPC approach to actual safeguards data streams. A significant step towards practical applicability of these technologies to authenticate, protect, and analyze actual safeguards data will be the development of an integrated platform that combines all methods. The paper will conclude with the current status of this development and an outlook on next steps. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525. SAND2021-1706 A.