Machine Learning Predicts Laboratory Earthquakes

Predicting the timing of an earthquake is a fundamental goal of scientists. A team from the University of Cambridge,  ‎Los Alamos National Laboratory and Boston University have identified a hidden sound signal emitted prior to a laboratory  earthquake. The experiment closely mimics real earthquakes. New developments in machine learning, a form of artificial intelligence, have been used to identify the tell-tale sounds emitted by the quake that predict precisely when an earthquake will occur. This is the first time that machine learning has been used to analyse the sounds emittied by an earquake prior to an eruption to predict when an earthquake will occur. The results could also be applied to predicting avalanches, landslides, failure of machine parts, and more.

Interestingly, the new research started in the GaN group in the Materials Department ‎ where Bertrand Rouet-Leduc, a PhD student of Colin Humphreys, was developing machine learning methods to make gallium nitride LEDs even more efficient. As part of the project, Bertrand visited Los Alamos National Laboratory in New Mexico where he met scientists working on earthquakes. He decided to see if his new machine learning methods could be applied to predicting if and when an earthquake would occur, and found out that they can do this very accurately. Bertrand has now been recruited by LANL to continue this work and apply it to real earthquakes.