E. Charniak, "Bayesian Networks Without Tears," AI Magazine, pp. 50--63, 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....uncertainty on the outcome reflecting the uncertainty inherent in the reasoning process itself. Ever since MYCIN [Shortliffe 76] the integration of uncertainty in medical expert systems has been a prime matter of concern. Currently, probabilistic methods such as Bayesian networks ( Rutledge 90b, Charniak 91, Berzuini 92, Heckerman 92] compete with possibilistic and fuzzy approaches. However, treatment of uncertainty seems not as widespread in diagnostic monitoring as it is in consultation systems: VM as one of the first, although employing the MYCIN formalism, did not make use of the inherited ....

E Charniak (1991) "Bayesian networks without tears", The AI Magazine 12(4), 50--63

....the number of causal connections and the number of hypothetical variables to the minimum at the expense of approximations in the representation of certain variable nodes. This is because the computational load increases by an order of 2 n 1 where n is the number of variable nodes in a network [13]. Third, in order to have efficient computation, it is computationally attractive to 4The x, y and width are measures in image coordinates. With a prc calibratcd camera and the geometry of perspective projection, they give corresponding three dimensional measures. 23 approximate any continuous ....

....measures. 23 approximate any continuous variable with a set of few discrete values. Fourth, the conditional prob ability distribution matrices between any two nodes are usually subject to probabilistic estimation based on extensive test examples. Statistical studies in the past [13] suggest that if a well controlled number of variables are built into a Bayesian network, the estimated distribution matrices capture the general characteristics of the problem. Accurate estimation of these parameters remains one of the important factors for the computational success of a belief ....

E. Charniak. "Bayesian networks without tears". AI Magazine, 12(4), 1991.

....to reasoning is based on soft, probabilistic decisions. Under such a framework all hypotheses are considered to some degree but with an associated probability. Bayesian Belief Networks provide a rigorous framework for combining semantic and sensor level reasoning under conditions of uncertainty [21, 6,8]. Given a set of variables W representing the scenario l, the assumption is that all our knowledge of the current state of affairs is encoded in the joint distribution of the variables conditioned on the existing evidence, P(w[e) Explicit modelling of this distribution is unintuitive and often ....

Eugene Charniak. Bayesian networks without tears. AI Magazine, 12(4):50-63, 1991.

....reasoning is based on soft, probabilistic decisions. Under such a framework all semantics are considered to some degree but with an associated probability. Bayesian Belief Networks (BNNs) provide a rigorous framework for combining semantic and sensor level reasoning under conditions of uncertainty [44,9,11]. Given a set of variables W representing the scenario 3, the assumption is that all our knowledge of the current state of affairs is encoded in the joint distribution of the variables conditioned on the existing evidence, P(wle ) Explicit modelling of this distribution is unintuitive and often ....

E. Charniak. Bayesian networks without tears. AI Magazine, 12(4):50-63, 1991.

....Decision Networks (DDN) A simpler chained structure with single causal dependencies over time, the Hidden Markov Model (HMM) is often used for speech analysis [7] and has been extensively adapted for analysis of dynamic scenes and perceptual control as described below. In probabilistic reasoning [8], the likelihood of classes of objects or events is inferred by propagation of belief values in the light of changing evidence. Early work incorporating Bayes nets was developed by Levitt and Binford [9] 10] to make model based vision reliable, while remaining computationally tractable. Bayes ....

E. Charniak, \Bayesian networks without tears," AI Magazine, vol. 12, no. 4, pp. 50-63, 1991.

....that the training data was generated by the model. The probability matrices are comprised by the transition probabilities, s, a, s # ) and the observation probabilities, # , a, z) Learning POMDPs can be performed by means of general algorithms for learning probabilistic networks from data [67, 41, 66, 7, 28, 57, 15]. This approach for learning POMDPs has been used until now in applications for speech recognition and dialogue systems. The Baum Welch learning rule, based on the expectation maximization (EM) algorithm, has also been applied to learning POMDP s in [8, 25, 20] for robotics applications. ....

Eugene Charniak. Bayesian networks without tears. AI Magazine, 12(4):50--63, 1991.

....since human experts have experience of application of uncertain rules, it should be much easier for them to give instances of uncertain reasonings. Incidentally, some researcher [5] thinks providing prior probabilities for root nodes is also a major diculty in constructing Bayesian networks [18, 3], which are currently very popular in many practical domains. In the PROSPECTOR model, supplying a subjective value for prior probability of the conclusion of every rule leads to an additional problem. That is, an inconsistency among values of prior probabilities may occur. To be consistent, this ....

....measures for uncertainties of a rule. These can satisfactorily capture the idea that the premise of the rule is a possible cause of the conclusion of the rule. Whereas those in the PROSPECTOR model fail to do so. Currently the most popular method for managing uncertainty is the Bayesian network [18, 3]. Probably this is principally because the model is consistent with probability theory. However, probabilistic inference using the Bayesian network is NP hard in general case [4] Clearly, the inference in our model is not NP hard, and our model is also consistent with Bayesian. There fore, we ....

Charniak, E., \Bayesian Network without Tears", The AI Magazine, 12:4, pp. 50-63, Winter 1991.

....the network with predecessors has an associated Conditional Prob ability Table (CPT) while nodes without predecessors have prior probabilities. The arcs in the networks define the dependencies amongst variables in the network, rendering nodes either dependent or independent given evidence (see [9] for a good explanation of indepen dence in Bayesian Networks) The network can be queried at any time to obtain the belief that a given node is a specified value. New evidence can be introduced into the network by setting the values of one or more of the variables in the network to a specific ....

E. Charniak. Bayesian networks without tears. AI Magazine, 12(4):50-63, 1991.

....classes of formula. While there is some connections between counting and computing probabilities, we know of know direct connections between these results and ours. An area in statistical reasoning which has been investigated from computational complexity point of is Bayesian belief networks (see [4, 16, 17] for information about them) Cooper [6] and Dagum and Luby [7] have shown that is is hard to even approximate the probabilities in such networks. 1.3 Approximate Consistency, Probability Intervals and Implied Correlations The Koller Megiddo result considers the standard, qualitative notion of ....

E. Charniak, Bayesian networks without tears, AI Magazine, 1991, pp. 50--63.

....turns on an outdoor light. However, she sometimes turns on this light if she is expecting a guest. Also, we have a dog. When nobody is home, the dog is put in the back yard. The same is true if the dog has bowel troubles. Finally, if the dog is in the backyard, I willprobablyhearherbarking. [6] . 2 4 3.1. A simple temporal aggregate, Vault Open, defined over four intervals. The conditional probability tables show Vault Open to be dependent on itself through some temporal causal relationship. 3 2 3.2. A probabilistic temporal network modeling a secure vault. ....

....turns on an outdoor light. However, she sometimes turns on this light if she is expecting a guest. Also, we have a dog. When nobody is home, the dog is put in the back yard. The same is true if the dog has bowel troubles. Finally, if the dog is in the backyard, I will probably hear her barking. [6] Bayesian networks are probabilistic intensional systems in which independence assumptions are used to restrict relevance. A Bayesian network is a directed acyclic graph (DAG) of random variable (RV) relationships. Directed arcs between RVs represent conditional dependencies. When all the parents ....

[Article contains additional citation context not shown here]

Charniak, Eugene. "Bayesian Networks Without Tears," AI Magazine, 12 (4):50--63 (1991).

.... or from observed temporal data (as modeled by Dynamic Bayesian networks) 3 (Static) Bayesian Networks Modeling Causality in Situations for Decision Theory Bayesian networks allow one to model situations in which causality plays a role but our understanding of what is going on is incomplete [Charniak91]. Networks that do not explicitly model temporal sequences, but rather causality relations (not necessarily temporal) between random variables, are generally referred to as (static) Bayesian networks. These variables may have discrete or continuous values, although most algorithms developed for ....

....related to decision theory. Here one specifies the desirability of various outcomes (their utility) and the costs of various actions that might be performed to affect the outcomes. The goal of decision theory is to find the action or plan that maximizes the expected utility minus the costs [Charniak91]. Bayesian networks extended for decision theory, by incorporating decision nodes and value nodes, are called influence diagrams. In Pathfinder [Heckerman90] a physician can manually enter information regarding a patient s symptoms and get the conditional probabilities of the diseases (related to ....

[Article contains additional citation context not shown here]

Charniak, Eugene. 1991. Bayesian Networks without Tears. AI Magazine, pp. 50-62.

....of a complex, non linear optimization program is the only method that can be generally adopted at this point. 6 EXAMPLE To illustrate the results and algorithms described previously, a simple example is discussed in this section. This example is based on the example described by Charniak [5] and on the calculations presented by Walley [28, Section 9.3.4] Consider the graph in Figure 1. There are five binary variables in the graph (the superscript c indicates negation) These relationships are summarized by the probabilistic model presented in Figure 1. Note that the probabilities ....

E. Charniak. Bayesian networks without tears. AI Magazine, pages 50--63, Fall 1991.

....all the remaining variables. These independence relationships can often be represented in terms of a graph, where the variables are associated with the nodes, and a missing edge represents a particular independence relationship (precise definitions can be found in the Appendix) See, for instance, [34, 29, 44, 12, 40, 11] for general reviews, treatments, or pointers to the large literature on this topic. The independence relationships result in the fundamental fact that the global high dimensional probability distribution P(x 1 ; xn ) over all variables, can be factored into a product of simpler local ....

....the marginals of P . 8. 3 The Directed Case: Bayesian Networks In the directed case, the family P(G) corresponds to the notions of Bayesian networks, belief networks, directed independence probabilistic networks, directed Markov fields, causal networks, influence diagrams, and even Markov meshes [34, 44, 12, 11, 22]. As already mentioned, the direction on the edges usually represents causality or time irreversibility. Such models are more common, for instance, in the design of expert systems. In the directed case, we have a directed graph G = V; E) The graph is also assumed to be acyclic, that is, with ....

[Article contains additional citation context not shown here]

E. Charniak. Bayesian networks without tears. AI Mag., 12:50--63, 1991.

No context found.

E. Charniak, "Bayesian Networks Without Tears," AI Magazine, pp. 50--63, 1991.

No context found.

Charniak, E. (1991). Bayesian networks without tears. AI Magazine, 12 (4), 50--63.

No context found.

E. Charniak, Bayesian networks without tears, AI Magazine 12 (4) (1991) 50--63.

No context found.

Charniak, E., "Bayesian Networks without Tears," AI Magazine , Vol. 12, no. 4, 1991, pp. 50-63.

No context found.

E. Charniak, "Bayesian network without tears," AI Mag., vol. 12, no. 4, 1991, pp. 50--63.

No context found.

E. Charniak. Bayesian networks without tears. AI Magazine 12 (1991), 4, 50-63.

No context found.

E. Charniak. Bayesian networks without tears. AI Magazine, 12(4):50--63, 1991.

No context found.

E. Charniak. Bayesian networks without tears. AI Magazine, 12(4):50-63, 1991.

No context found.

Eugene Charniak. Bayesian networks without tears. AI Magazine, 12(4):50-63, Winter 1991.

No context found.

E. Charniak. Bayesian networks without tears. AI Magazine, 12(4):50--63, 1991.

No context found.

Eugene Charniak. Bayesian networks without tears. The AI Magazine, 12(4):50--63, 1991. 39

No context found.

E. Charniak. Bayesian networks without tears. AI Magazine, 12(4):50--63, 1991.

First 50 documents  Next 50

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