The network architecture is a balanced binary
tree. The only requirement is for a broadcast and
gather facility. The tree starts empty and dynamically
adapts to new knowledge. During learning, nodes are
added and deleted. When a node is added, its position
is determined so as to maximize breadth. When a
node is deleted, transformations are applied to keep
the tree balanced. Hence, the tree is always
maintained as full as possible (see [18]).