I'll start with a long and detailed example of what the dynamic networks can be used for. Then explain how to accomplish it with this important and needed framework. (Something similar exists on a higher level in D3.js but AFAIK nothing specific for network self-automation).
The basic networks of
this example are constructed from a written text, where each word is linked to its sentence node, and the sentences are linked to each other in the paragraph.
The sequences of words can also be networks alongside the sentences network. And so can the sentence sequences.
A grammar network would "drink in" the basic sentences and create a new network of verbs and object phrases, with verb-based utterances linking to the object phrases through different grammatical connections. So now you have "a lexicon" of the words in your text. This can be derived from the context and order of the words in the sentences, or by formerly learned and known word meanings in a dictionary (which can be looked up). This network may have "options" not decided yet, of the grammatical part of the words. For example, the phrase "drink in" is a verb, but the word drink may be an object. At an early stage of processing, the word "drink" would be in two nodes, one in the paragraph's verb network, and another, in the objects network.
A representation network would break down the object phrases into basic nodes with the important object and its extra descriptions.
A structural-semantic network would "drink in" the grammar network and spit out an equivalent of the semantic structure without the actual words.
A network of semantic understanding would connect the semantic structure network to the verbs and objects in the grammar network.
A cognition network would add expected responses such as feelings, actions, further thoughts, memories, and information storage.
Finally, a response network could build several possible responses building on the cognition network alongside the other networks.
The temporary networks cause the arousal of several possible versions for each of the options of comprehension, and by weighing in on certain connections we get a comprehensive result and response. These decisions are also stored in a decision network storing the considered priorities and the history of changes in time.
All this is possible if we can build a framework for automatically and dynamically connecting and disconnecting nodes in a "graph" and changing their weights through a comprehensive rule based