Now imagine taking the 4×4 image obtained after the first convolution and applying another convolution using a different 2×2 kernel. The output will be a 3×3 image. This will require 9 neurons. This is the second hidden convolution layer. Each neuron in this layer is the equivalent of a complex cell in Hubel and Wiesel’s hierarchy. Each neuron in this layer is sensitive to the value of 4 pixels in the 4×4 image generated by the previous layer.
