How to Identify Patterns of Citywide Dynamic Traffic at a Low Cost? An In-Depth Neural Network Approach with Digital Maps

Figure 3

Parameters of the proposed stacked convolutional autoencoder. There is a fully connected layer with eight neurons and a softmax classifier with three neurons added at the end of the encoder. For the eight convolution layers in the encoder and the convolution layers from 9 to 16 in the decoder, kernel size, activation function, padding, and max-pooling are the same, only the number of kernels is changing, shown in the encoded and decoded part, respectively. For the convolutional layer 17, the number of kernels is 3, and the activation function is sigmoid.