Talk:URE historical discussion

From OpenCog
(Redirected from Talk:Unified rule engine)
Jump to: navigation, search

Some brainstorms I had recently:

It seems to me that Rete networks are just another network..

I wondered if it would not be useful to implement a naive system as a network, but use a neural network to decide which precedents to evaluate. A Rete adds links from rule nodes to fact nodes or other rule nodes. Since you already have a hypergraph, why not back-propagate some weight to links which jump over intermediate logic to allow the rule's dependency digraph to become incrementally shallower with each evaluation? Save memory by occasionally pruning weak links.

If you treat nodes along multiple levels of dependency as a vector, that vector (or its hash) might be an ID for a node with its own links. Aphor 21:55, 29 April 2010 (CDT)