Radiation threat detection is an important security challenge for many countries. This paper introduces Canonical Autocorrelation Analysis (CAA), a method for finding multiple-to-multiple linear correlations within a single set of features. This can be useful when looking for hidden parsimonious structures in data, each involving only a small subset of all features. The discovered correlations are highly interpretable as they are formed by pairs of sparse linear combinations of the original features. In addition, this results in useful visualizations of data.
In this paper it is shown how CAA can be of use as a tool for anomaly detection when an expected structure of correlations is not followed by anomalous data. In the case of radiation threat detection, this allows characterization of harmless radiation data based on the patterns across bins of photon counts, and flagging of anomalies where they fail to follow such patterns. The resulting technique performs significantly better than an unsupervised alternative prevalent in the domain, while providing valuable additional insights for threat analysis.
Artur Dubrawski (Advisor)
Simon Labov (Lawrence Livermore National Laboratory)