H.L. Chin and G.F. Cooper, "Bayesian Belief Networks Inference Using Simulation," L.N. Kanal, T.S. Levitt, and J.F. Lemmer, eds., Elsevier Science, 1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....prior probability of the evidence is usually very small, and thus logic sampling performs poorly. Two other papers enhance logic sampling by examining evidential integration, which employs arc reversal to evidence nodes that are sinks to sources to avoid the computational penalty of observed nodes [FC89, CC89]. Likelihood weighting (LW) or evidence weighting are designed to get around the problem of logic sampling [FC89, SP90] In likelihood weighting, every time we reach an evidence node, we don t sample and throw away inconsistent samples; instead we take the observed value of the evidence variable, ....

H. L. Chin and G. F. Cooper. Bayesian belief network inference using simulation. In L. N. Kanal, J. F. Lemmer, and T. S. Levitt, editors, Uncertainty in Artificial Intelligence 3, pages 129--148. North Holland, New York, 1989.

....as the fraction of simulations that give rise to the observed set of evidence. This method is linear in the number of nodes in the network, regardless of the degree of interconnectedness of cycles. Unfortunately, it is exponential in the number of pieces of evidence observed. Chin and Cooper [18] have used the logic sampling approach to generate samples of medical cases for simulation purposes. They avoid the exponential complexity of the general problem by rearranging the direction of the links in the network using Shachter s algorithm, so that all observed variables are inputs (sources) ....

H.L. Chin and G.F. Cooper. Bayesian belief network inference using simulation. In L.N. Kanal, J.F. Lemmer, and T.S. Levitt, editors, Uncertainty in Artificial Intelligence 3, pages 129--148. North Holland, New York, 1989.

....reducing the error that can be used with either LW or LS [25, 15] LW and LS both belong to a class of stochastic simulation methods, called forward propagation, because values are first assigned to root nodes, then propagated towards the leaves along the direction of the arcs. Evidence Reversal [4] is a technique in which evidence nodes are turned into root nodes by reversing any incoming arcs and recomputing the appropriate conditional probabilities. The advantage of this process is that it produces a version of the network which is very suitable for forward propagation. Unfortunately, if ....

....with zero, and using strategies for zero compression [16] and outlines cases in which each technique should be preferred. Node removal, node merging and network pruning are some other ways of obtaining an approximate model, simpler than the original network. These methods are described in [4]. 5 Domain Characterisation We characterise a problem by obtaining measurements for a set of features of the network, individual nodes, the set of instantiated nodes (evidence) and the set of queried nodes (targets) These measurements are taken prior to each experiment and are used later for ....

Homer L. Chin and Gregory F. Cooper. Bayesian belief network inference using simulation. In Uncertainty in Artificial Intelligence 3, pages 129--147, 1989.

....by changing the size of the evidence set E , i.e. by changing the number of clamped variables. The size of the configuration space was increased in two ways: by allowing more variables in the Bayesian networks, and by allowing the variables to have more values. 6.2. The Results It has been noted [61] that solving certain types of probabilistic reasoning tasks can become very difficult if the Bayesian network contains a lot of extreme probabilities (probabilities with values near zero or one) However, as already noted in [38] in our MAP problem framework this does not seem to be true. We ....

H.L. Chin and G.F. Cooper. Bayesian belief network inference using simulation. In L.N. Kanal and J.F. Lemmer, editors, Uncertainty in Artificial Intelligence 3, pages 129--147. Elsevier Science Publishers B.V. (North-Holland), Amsterdam, 1989.

....into a tree structure with a query node at the root of the tree. Estimates for the belief of the query node are iteratively computed by traversing the tree. This incremental evaluation results in a form of anytime algorithm. Other researchers have developed anytime algorithms for BN evaluation [5, 19, 13] and proposed the use of tree representations for efficient representation of evaluation of a BN [8, 4] Sect. 2) in this paper we combine these approaches. The representation, construction and tree traversal evaluation of a treeNet is described in Sect. 3. A best first tree traversal ....

....of the result is measured in terms of the distance between the exact and estimated beliefs of query nodes. If this distance decreases as computation time increases, the algorithm has an anytime flavour. The rate at which this distance decreases is obviously important. Simulation approaches [5] are classified as anytime algorithms because the accuracy of results improves as the sample size grows. Other anytime algorithms for BN evaluation have been developed [13, 19] and compared [16] In Wellman and Liu s state Space abstraction [19] the states of selected nodes are merged; the ....

Homer L. Chin and Gregory F. Cooper. Bayesian belief network inference using simulation. In Uncertainty in Artificial Intelligence 3, pages 129--147, 1989.

....distribution converges on a probability with additional computation depends on the topology of the network, and on the nature of the probabilistic dependencies within the network. Recent work has shown current simulation algorithms to have intolerably slow convergence rates in many realistic cases [4] . Stochastic simulation is nevertheless a promising class of inference for the derivation of useful bounded resource computation strategies. ffl Completeness Modulation Completeness modulation strategies center on techniques for reasoning about attributes of the uncertain reasoning model to ....

H.L. Chin and G.F. Cooper. Bayesian belief network inference using simulation. In L.N. Kanal, J.F. Lemmer, and T.S. Levitt, editors, Uncertainty in Artificial Intelligence 3, pages 129--148. North Holland, New York, 1989.

....acyclic graph (dag) as often done with an exact approach. In the stochastic methods (e.g. Henrion s logic sampling method [16] and Pearl s stochastic simulation method [21] the precision of the reasoning is dependent on the size of stochastic samples that a simulation generates. Chin and Cooper [3] have noted that the stochastic simulation algorithms, when applied to certain networks, could lead to much slower than expected convergence to the true posterior probabilities, and proposed several possible forms of graph modification. All the above mentioned evidential reasoning methods rely ....

Homer L. Chin and Gregory F. Cooper. Bayesian belief networks inference using simulation. In L. N. Kanal, T. S. Levitt, and J. F. Lemmer, editors, Uncertainty in Artificial Intelligence, pages 129--147. Elsevier Science Publishers B. V. (NorthHolland) , 1989.

....approach to complexity reduction is to approximate the model by preprocessing the network. Methods have been proposed which do this by one of the following methods: removal of weak links [10] state space abstraction [12] replacing small probabilities with zero, graph pruning and node merging [2]. These procedures may be applied individually or in combination. It is possible to use such procedures as anytime algorithms to obtain results that improve as more time, memory or information are available. For example we can iteratively replace deleted weak edges and re evaluate the network. If ....

Homer L. Chin and Gregory F. Cooper. Bayesian belief network inference using simulation. In Uncertainty in Artificial Intelligence 3, pages 129--147, 1989.

....successive instantiations of nodes in Markov sampling are not independent. The method s convergence rate may therefore deteriorate when the network contains links that are near deterministic because the network may get trapped in scenarios from which it takes many instantiations to escape [128, 30, 28]. In addition, the computation per trial is often greater than that needed for straight logic sampling because the pre processing step must take place for every node and for every trial. Several other approaches have been proposed as ways of overcoming logic sampling s difficulties with actual ....

....Several other approaches have been proposed as ways of overcoming logic sampling s difficulties with actual observations. Evidence integration reverses arcs (by iterative applications of Bayes rule) to convert nodes whose values have been observed into sources (i.e. all arcs emanate outward)[30]. This technique works where it is applicable, but remains general case intractable. In likelihood weighting or evidence weighting, posed independently by Fung and Chang [51] and by Shachter and Peot [128] respectively, nodes representing variables that have already been observed are not ....

[Article contains additional citation context not shown here]

H. L. Chin and G. F. Cooper. Bayesian belief network inference using simulation. In T. S. Levitt L. N. Kanal and J. F. Lemmer, editors, Uncertainty in Artificial Intelligence 3, pages 129--147. North Holland, Amsterdam, 1989.

No context found.

H.L. Chin and G.F. Cooper, "Bayesian Belief Networks Inference Using Simulation," L.N. Kanal, T.S. Levitt, and J.F. Lemmer, eds., Elsevier Science, 1989.

No context found.

H.L. Chin and G.F. Cooper. Bayesian belief network inference using simulation. In L.N. Kanal and J.F. Lemmer, editors, Uncertainty in Artificial Intelligence 3, pages 129--147. Elsevier Science Publishers B.V. (North-Holland), Amsterdam, 1989.

No context found.

H.L. Chin and G.F. Cooper. Bayesian belief network inference using simulation. In L.N. Kanal and J.F. Lemmer, editors, Uncertainty in Artificial Intelligence 3, pages 129--147. Elsevier Science Publishers B.V. (North-Holland), Amsterdam, 1989.

No context found.

H. L. Chin and G. F. Cooper. Bayesian belief network inference using simulation. In L.N. Kanal, J.F. Lemmer, and T.S. Levitt, editors, Uncertainty in Artificial Intelligence 3, pages 129--148. North Holland, New York, 1989.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC