Predicting the probability of an event occurrence with machine learning models often requires a large amount of sample data for training.
However, rare catastrophic events such as the New Crown pandemic, California mountain fires, and tsunamis lack historical data to predict their next probability of occurrence by machines.
A recent study in Nature Computational Science proposes a way to train models to predict rare events with a small amount of data.
Scientists have introduced active learning algorithms into the DeepOnet model.
Active learning is able to prioritize data sets, train supervised models by analyzing small amounts of data and labeling the remaining data.
DeepOnet’s two parallel neural networks, on the other hand, are able to quickly analyze large amounts of data and scenarios and output probability sets.
The combination of the two enables the model to actively search with few data points and identify precursors to rare events.
The new method outperforms traditional models in predicting the probability that a ship will be hit by a large wave and break apart.
and it is expected to be used in the future to predict extreme weather events such as hurricanes.