Example of an autoencoder neural network. Each column represents a neuron layer. Each arrow represents a connection between the neuron layers. The first and last layers are composed from the measured and denoised intensities of a spectrum at the different channels, respectively. The bottleneck contains the minimum number of neurons needed to compress the data without loss of signal information. In this example, the signal intrinsic dimension (size of the bottleneck) is three while the data extrinsic one (size of the input and output spectra) is 10.