Advances in computing, algorithms, and new data sources present significant opportunities to leverage cutting edge data analytics in support of nonproliferation and arms control. The promises of data analytics are of accelerated analyst workflows, exploitation of vast troves of open-source information, and automated fusion and discovery across distinct data modalities. However, off-the-shelf analytics alone are insufficient to realize these promises; the high-consequence nature of non-proliferation applications requires the next generation of advanced data analytics techniques. This application space is characterized by complex systems and noisy data, sparsity and rare events, and the need for robust decision support under uncertainty. Addressing these challenges requires concerted and intentional design into data analytics systems. Domain aware analytics can rely on a much wider swath of information than available data to constrain approaches and make predictions. Techniques that are explainable, interpretable, and transparent are more accessible and directly useful for intended end users and decision makers. Approaches that perform reliably and predictably can be instilled with greater trust in dynamic and often adversarial contexts. Analytics that are highly specialized to the unique applications and data of nonproliferation can best represent the precise signatures, patterns, and behaviors of interest. This session will highlight research that explores the challenges of the nonproliferation application space and seeks to develop the next generation of data analytics approaches in response. This work was funded by the U.S. Department of Energy National Nuclear Security Administration’s Office of Defense Nuclear Nonproliferation Research and Development (NA-22). SAND2022-2653 A.
Year
2022
Abstract