Deep learning for aftershock location patterns and the earthquake cycle

Phoebe DeVries, Thomas B. Thompson, Martin Wattenberg, Fernanda Viegas, & Brendan J. Meade

Published August 2, 2018, SCEC Contribution #8251, 2018 SCEC Annual Meeting Talk on Tue 13:30

Over the past few years, deep learning has led to rapid advances in applied computer science, from machine vision to natural language processing. These methods are now accessible to scientists across all disciplines due to the availability of easy-to-use APIs and affordable GPU acceleration. We demonstrate two specific applications of deep learning within earthquake science. In the first, we train a deep neural network to learn computationally efficient representations of viscoelastic solutions, across large ranges of times, locations, and rheological structures. Once found, these efficient neural network representations may accelerate computationally intensive viscoelastic calculations by more than 50,000%. In the second, we focus on aftershock location patterns and find that a fully connected neural network trained on 131,000+ mainshock-aftershock pairs can explain aftershock locations in an independent testing data set of 30,000+ mainshock-aftershock pairs more accurately (AUC = 0.849) than static elastic Coulomb failure stress change (AUC = 0.583). In contrast to the common assertion that deep learning produces “black box” results, in both applications, the trained neural networks can provide unique physical insights.

Citation
DeVries, P., Thompson, T. B., Wattenberg, M., Viegas, F., & Meade, B. J. (2018, 08). Deep learning for aftershock location patterns and the earthquake cycle. Oral Presentation at 2018 SCEC Annual Meeting.

Related Projects & Working Groups
Computational Science (CS)