Signal Matching via Computationally Efficient Hashing

Vadum will implement and test a software capability for rapid matching of time series data to a database of previously observed time series. The Clustered, Shingled Hashing of Random Projections (CSHARP) capability will generate efficient fingerprint representations of time series data. This capability will enable rapidly query for matches to newly observed, high-volume and velocity signal streams. CSHARP will provide an efficient alternative to current Normalized Cross Correlation (NCC) methods, retrieving accurate match sets in at least an order of magnitude less time, with acceptable loss in precision. CSHARP will provide a confidence on returns associated with a query, and will detect incoming data stream statistical drift assisting human-in-the-loop mitigations. The CSHARP capability, and associated graphical user interface, will be integrated into existing NGA analyst work flows, enabling scaling to dynamic, increasing GEOINT activities.