Application of Neural Networks to Nuclear Safeguards

With the increase in availability of vast amounts of data and the likelihood that data is only going to accumulate and grow in the future, conventional methods that relied on scarce data for interpretation and inference of physical phenomena are rapidly losing popularity in many fields of science and engineering. This has paved the way for the debut and establishment of novel methods of machine learning such as artificial neural networks, Gaussian processes and the like as a means to build surrogate models for complex physical phenomena instead of costly, complicated, and time-consuming simulations. In this report, we give a broad survey of utility and progress in terms of applications of such modeling practices in the field of nuclear safeguards. We also present the results of a short scoping performance evaluation of a deep artificial neural network. To achieve this, the popular open source machine learning library named tensorflow-keras will be used along with numerical computing and visualisation tools such as Pandas, NumPy, and Matplotlib