Fig. 4.

A regular convolutional neural network (CNN). The input consists of S samples and P features. The 1D filter with kernel size of K and L channels is used for convolving data with the input. By pooling (downsampling) with kernel size of 2, the resulting tensor now becomes approximately of size S×P/4×L. The fully connected layer considers all the features in every channels and output the probability of class labels (C) for each sample