Fig. 5: Evolving friction time series predictions from convolutional encoder–decoder (CED) models. | Nature Communications

Fig. 5: Evolving friction time series predictions from convolutional encoder–decoder (CED) models.

From: Predicting fault slip via transfer learning

Fig. 5

a Model trained and tested on the laboratory data. b Model trained on simulations and tested on laboratory data. The first 20% of the acoustic emission (AE) signal (0–60 s) was used for training to construct the model. c Cross-trained model. Model trained on the simulations, then fixing the encoder and decoder layers, with the model then additionally trained on the bottleneck (latent space) applying a portion of the laboratory (experiment p4677) data. The mean absolute percentage error (MAPE) is listed for each prediction. See “Methods” section for details.

Back to article page