Schematic of a ConvNet. This network is a hybrid between a set of convolutional layers and a fully connected feed forward network. The convolutional layers contain kernels that scan through the input and extract feature maps (vertical black solid lines). Each layer forms successively more abstract and sophisticated maps, with each filter being sensitive to a particular feature that the network deems important for the classification task. The high level features which the network automatically generates are then passed through a max pooling layer (gray vertical lines), which take summary statistics of the maps, reducing the number of parameters and promoting generalization. The final content is then fed into the fully connected layers which behave as described in Fig. A.1, and are responsible for classifying the spectrum into either the AR or PF class. It is important to note that the weights and kernels are updated during training, meaning that the ConvNet adaptively searches for a set of features that is most helpful for distinguishing between AR and PF spectra.