Signal Matching via Computationally Efficient Hashing

Vadum will develop and demonstrate a software tool for rapid, accurate matching of time-series data to a previously collected database. The Clustered, Shingled Hashing of Random Projections (C-SHARP) tool maps a library of time series data to a compact database of fingerprints. The tool enables efficient queries for matches to a new time series by identifying fingerprint similarity. To support different statistical characteristics of data sources, the tool applies a sampling based self-configuration procedure that selects from a suite of hashing techniques, configures hyperparameters, and applies a dictionary learning procedure to optimize a matching objective. C-SHARP queries achieve high Top-N set overlap with current techniques based on Normalized Cross Correlation (NCC) , particularly when paired with a secondary NCC filtering stage that operates on an initial candidate set. CSHARP demonstrates an order of magnitude speed improvement over NCC queries, even when paired with a secondary NCC filter.