Home
The capacity of our RNN models to extract multiple target waveforms, which are non-linear transformations of the input, may be tested on simulated data. To this end we trained RNNs to predict the phase of either one of the two sinusoidal input signals representing undersampled cardiac and respiratory waveforms plus Gaussian noise.
The figure below illustrates that the target waveforms (dotted lines) were predicted well (solid lines) from the undersampled input data (solid dots), as long as the input noise was low (y-axis: SD ranging from 0.0 to 1.0).
