How Google and Harvard Designed AI To Predict Earthquake After-Shocks

Now if AI could only prevent earthquakes...JL

Johnny Lieu reports in Mashable:

Researchers trained a neural network on 131,000 mainshock–aftershock pairs, then selected unrelated 30,000 pairs for a test. They found the artificial intelligence system was more reliable at detecting the locations of aftershocks than a widely-used model for predictions. "Aftershock forecasting is well-suited to machine learning because there are so many physical phenomena that could influence behavior and machine learning is extremely good at teasing out those relationships."
After the shock of an earthquake, aftershocks can wreak serious havoc too.
These subsequent quakes can last for weeks and months, and while scientists have developed models to try to detect their size and timing, figuring out where these aftershocks will hit is more difficult.
In a study published in Nature by a team from Google and Harvard, deep learning was utilized to help predict where these aftershocks hit.
The researchers trained a neural network on more than 131,000 mainshock–aftershock pairs, then selected unrelated 30,000 pairs for a test. In their test, they found that the artificial intelligence system was more reliable at detecting the locations of aftershocks than a widely-used model for predictions, known as the Coulomb failure stress change.
The network also came up with an effective prediction metric that hadn't been popular in earthquake science, but rather in metallurgy, or the study of metals.
"There are three things you want to know about earthquakes — you want to know when they are going to occur, how big they're going to be and where they're going to be," Brendan Meade, Harvard's Professor of Earth and Planetary Sciences, told ScienceDaily.
"Prior to this work we had empirical laws for when they would occur and how big they were going to be, and now we're working the third leg, where they might occur."
Meade came up with the idea of using artificially intelligent neural networks on aftershocks a few years ago. His next step is to try to predict the magnitude of earthquakes with the help of AI.
"I'm very excited for the potential for machine learning going forward with these kind of problems — it's a very important problem to go after," Phoebe DeVries, a co-author of the study, told ScienceDaily.
"Aftershock forecasting in particular is a challenge that's well-suited to machine learning because there are so many physical phenomena that could influence aftershock behavior and machine learning is extremely good at teasing out those relationships. I think we've really just scratched the surface of what could be done with aftershock forecasting ... and that's really exciting."
While the system is still imprecise, and thus a long way off from being used officially, it's certainly a promising start.