They found that their network was able to identify and explain the pattern of aftershock locations in an independent dataset of more than 30,000 earthquake and aftershock pairs more accurately than Coulomb failure stress change could.
This, they argued, highlighted how deep learning approaches could lead to improved aftershock forecasts, while providing insights into the mechanisms of earthquake triggering.
If so, the approach could have big benefits in New Zealand, where thousands of aftershocks have been recorded in the wake of the 2010 7.1 Darfield quake and 2016's 7.8 Kaikoura Earthquake.
The University of Otago's chair of earthquake science, Professor Mark Stirling, said Coulomb stress change had been investigated as a forecasting tool for decades and had a "solid physical basis".
But there were cases where it hadn't proven as effective - such as work on the Canterbury earthquake sequence - and other clustering-based methods had been much more informative.
Stirling thought the new research was worthwhile as it looked more deeply into how the Coulomb-based forecasting method could be properly developed and tested.
"The application of machine learning to high-quality earthquake datasets is a big step beyond what has been done in the past," he said.
"With evolving methods like this, we stand to gain a better understanding of how this method can contribute to the ensemble of existing earthquake forecasting methods."
GNS Science experts Dr Matt Gerstenberger, Dr David Rhoades and Dr Bill Fry told the NZ Science Media Centre the new technology would be worth watching.
"The spatial patterns identified by the machine learning are consistent with those used in statistical models for earthquake forecasting in New Zealand, and elsewhere around the world," they said.
"It will be interesting to see if machine learning will be able to identify spatial patterns that will help to improve traditional forecasting in the future."
