treeNets: A framework for anytime evaluation of belief networks

We present a new framework for evaluation of belief networks (BNs). It consists of two steps: (1) transforming a belief network into a tree structure called a treeNet (2) performing anytime inference by searching the treeNet. The root of the treeNet represents the query node. Whenever new evidence is incorporated, the posterior probability of the query node is re-calculated, using a variation of the polytree message-passing algorithm. The treeNet framework is geared towards anytime evaluation. Evaluating the treeNet is a tree search problem and we investigate different tree search strategies. Using a best-first method, we can to increase the rate of convergence of the anytime result.