Fig. 1: Autoencoder’s learned representation.
From: Deep multi-task learning for early warnings of dust events implemented for the Middle East
a The code’s regional task representation: Selected examples from the validation set of original and predicted regional PM10 fields. Prediction was performed using the model’s autoencoder, \({\Phi}_{reg}\). b Code’s local task representation: Boxplots of the first PCA component of the encoder codes over eight time points in 12 h intervals. Sequences that are followed by dust events in 24 h after the last time points (orange), and those that do not (blue) are compared.
