5.1.2. Step 2. Number of convolutional filters
In this step, the goal is to minimize the width of convolutional layers (here, the width refers to a number of convolutional filters at each convolutional layer).
As detailed in Table 3, the number of convolutional filters of the CNN D is divided by 2 comparing to the CNN B (which has been selected during the first optimization step). The results obtained using CNNs B and D are summarized in Fig. 3. Fig.
The CNN D is clearly below the selection threshold on the validation set (with respect to standard deviations). Therefore, the CNN B is selected again after the second optimization step.