Hacker News Reader

I would love to implement a graph like this where each node used bayesian inference to act on messages received from connected vertices.

My gut says if you could get the system's state to settle to an equilibrium, it could react to changes in external signals in a probabilistic way and learn.

Instead I'm busy working on an iphone app...

View on HN

Check out en.wikipedia.org/wiki/Junction_tree_algorithm not a great article, but the referenced literature is ok

Don't have much time to elaborate at the moment, but look up the "junction tree algorithm"- it's a way of performing inference in graph-structured statistical models. You think of edges as relationships between random variable (which are the nodes), and have the nodes communicate with each other until all the signals have propogated. Makes inference straightforward, though still exponential

http://en.wikipedia.org/wiki/Bayesian_network

Reply on news.ycombinator.com