I had a dog named Murphy that could predict earthquakes. About an hour before they happened, Murphy would grow agitated—crazed, warning me that something was coming.
Researchers from Brown and MIT are teaming up to use an advanced ML system and sequential sampling techniques to forecast extreme events such as earthquakes. Predicting disasters like earthquakes, pandemics, or “rogue waves” is no small feat. Computational modeling faces an almost insurmountable challenge: these events are so rare that we don’t have enough data for predictive models to forecast them accurately.
But, a team of researchers from Brown University and Massachusetts Institute of Technology says it doesn’t have to be that way.
In a new study in Nature Computational Science, researchers from Brown University and MIT describe how they combined statistical algorithms, which need less data for accurate, efficient predictions, with a powerful ML technique developed at Brown, and trained it to predict scenarios, probabilities, and timelines of rare events despite the lack of historical records on them. They used a sequential sampling technique called active learning. These types of statistical algorithms can learn to label new relevant data points that are equally, or even more important, to the outcome they’re calculating. At the most basic level, they allow more to be done with less.
That’s critical to the ML model the researchers used in the study. Called DeepOnet, the model is a type of artificial neural network using interconnected nodes in successive layers that mimic the connections made by neurons in the human brain. DeepOnet is more advanced and powerful than typical artificial neural networks because it’s two neural networks in one, processing data in two parallel networks.
The research team shows that, combined with active learning techniques, they can train the DeepOnet model on the parameters or precursors that lead up to the disastrous event someone is analyzing, even when there are not many data points.
The researchers found that their new method outperformed more traditional modeling efforts and that it presents a framework that can efficiently discover and predict rare events.