Why is the 3rd layer not fully connected in this example?

Question

Neural networks

Here is an interactive example of the work of a convolutional neural network: scs.ryerson.ca/~aharley/vis/conv
Why in this example the 3rd layer is not fully connected and how did you determine which feature matrices to connect with each other and which not?

Answer 1

Maybe they did Dropout so as not to retrain https://www.cs.toronto.edu/~hinton/absps/JMLRdropo...

Answer 2

This is "Downsampling layer 1" aka pooling:
https://en.wikipedia.org/wiki/Convolutional_neural...