Fig. 7: Prediction of spectral correlation maps by artificial neural networks trained from experimental datasets.
From: Deep learning prediction of noise-driven nonlinear instabilities in fibre optics
a Two-seed experiment: Example of two experimental correlation maps predicted from the neural network for different seeding conditions. The correlation maps predicted by the artificial neural network (ANN) trained either from the input seed parameters (bottom panels) or from the output spectrum (middle panels), are compared to the correlation maps retrieved from experiments displayed on the bottom panel (top panels). b Same as panel (a), illustrating two examples of correlation maps predicted by the ANN for experiments using four input seeds. In all cases, the root mean square error (RMSE) value displayed is computed as the average of each pixel RMSE over the whole map.
