Optimizing CNN parameters

CNNs have many different parameters. Training a CNN model requires many input images and performing processing, which can be very time-consuming. If the parameters that are selected are not optimal, the whole process has to be repeated again. This is why it is important to understand the functions of each parameter and their interrelationship: so that their values can be optimized before running the CNN to minimize repeat runs. The parameters of a CNN are as follows:

• Image size = (n x n)
• Filter = (f_h,f_w), f_h = filter applied to image height, f_w = filter applied to image width
• Number of filters = n_f
• Padding = p
• Stride = s
• Output size = {(n + 2p - f)/s +1} x {(n + 2p - f)/s + 1}
• Number of parameters = (f_h x f_w + 1)x n_f

The critical task is to select the aforementioned parameters (filter size (f), number of filters (nf), stride (s) in each layer, padding...