Data Discovery and Anomaly Detection Using Atypicality: Signal\n Processing Methods

The aim of atypicality is to extract small, rare, unusual and interesting\npieces out of big data. This complements statistics about typical data to give\ninsight into data. In order to find such "interesting" parts of data, universal\napproaches are required, since it is not known in advance what we are looking\nfor. We therefore base the atypicality criterion on codelength. In a prior\npaper we developed the methodology for discrete-valued data, and the the\ncurrent paper extends this to real-valued data. This is done by using minimum\ndescription length (MDL). We show that this shares a number of theoretical\nproperties with the discrete-valued case. We develop the methodology for a\nnumber of "universal" signal processing models, and finally apply them to\nrecorded hydrophone data.\n