Seismic networks are deployed locally and globally, and record continuous data that is now permanently archived. These data contain rich information about earthquake processes, but standard processing methods leave much information unused.  

Seismology is fortunate to have large, manually labeled data sets, which provide an excellent opportunity for developing deep learning algorithms for earthquake monitoring.  Deep learning is an effective, data-driven way to identify a nonlinear map from a high-dimensional input distribution to a target distribution of interest. 

In this work, we present our recent developments in developing deep learning methods to denoise, detect, pick, and associate earthquakes, which lead to improved earthquake catalogs. Deep learning techniques are rapidly advancing, and seismology is poised to improve earthquake catalogs dramatically, which will provide a much clearer and more detailed picture of earthquake processes.