Extended Data Fig. 1: Comparison of spectral intensity evolution modelling by the RNN when trained using data input in linear and logarithmic (dB) representation.
From: Predicting ultrafast nonlinear dynamics in fibre optics with a recurrent neural network

The dashed red line shows the simulated spectra of a higher-order soliton (N = 6) at selected distances along the fibre similar to those plotted in Fig. 5. The solid blue lines shows the RNN prediction when trained directly with spectral data input in dB. The solid green line shows the RNN prediction (plotted in dB) when trained with spectral data input in linear scale.