Template matching methods have great sensitivity for detection of seismic events, but are dependent on the availability of suitable template earthquakes. We show here that an alternative method, the matrix profile (MP), recently developed in the computer science domain, is effective for seismic event detection. The MP method does not require {\it a priori} templates; instead, it efficiently calculates the maximum autocorrelation of all subsequences of a seismic waveform. Seismic events have much higher similarity to each other than to background noise, and thus portions of the waveform with high MP values are likely seismic events.
We apply the MP to three seismic case studies – local event detection at Parkfield, CA, teleseismic event detection with the global seismic network, and laboratory experiment data. We demonstrate that the MP can detect events that are present in earthquake catalogs, but also a significantly higher number of events (up to $10\times$) that were not routinely identified. We show that the MP can be used to pick phase onsets, measure event durations, and identify novel events in seismic data that can be used as templates for future analyses.