Researchers at Rice University trained a deep neural network to recognize complex flows of air or water and predict how they will change over time. This image depicts the significant differences in the scale of features shown to the model during training [top] and the features it learns to recognize [bottom] to make predictions.