Heckerman, D., J. S. Breese, and Koos Rommelse, "Decision Theoretic Troubleshooting," Communications of the ACM, Vol. 38, No. 3, pp. 49-56, March 1995.  Home/Search   Document Details and Download   Summary   ACM   TOC   Related Articles   Check

....with Bayesian networks applied to adaptive diagnosis and tutoring. 3 Troubleshooting Troubleshooting a man made device is often quite a complex task. Therefore a system that uses evidence derived by performing repair actions or observations can substantially shorten the troubleshooting process [6]. 3.1 Bayesian network model for troubleshooting Within the approach presented in [10] a troubleshooting problem is modelled with a Bayesian network encoding relations among three types of variables: faults of the device F # F , actions A # A troubleshooting steps that may fix the problem, ....

....Bayesian network models. Since the models have the structure of the Nave Bayes model, many probability propagations can be performed whenever a new troubleshooting step has to be chosen. The approach exploits heuristics based on the P (A = yes) c A ratio and extends the myopic approach proposed in [6] for troubleshooting that includes observations. In [10] the strategies obtained using the SACSO approach were compared with the optimal solutions. The ECR values of the strategies provided by the SACSO troubleshooter were very close to the optimal values. The troubleshooter is available on the ....

D. Heckerman, J. S. Breese, and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, 38(3):49--57, 1995.

....task is to nd a troubleshooting such that for all possible strategies s it holds that ) ECR(s) 1. 3 Various setups of troubleshooting It is often reasonable to assume that, if the device is malfunctioning, then there is only one fault in the device (single fault assumption) [3]. Otherwise the situation is referred to as multiple faults. When the faults are independent, i.e. p(F 1 ; F jFj ) F i 2F p(F i ) we speak about independent faults [10] A solution of the troubleshooting task can be easily found in the case of independent actions, that is in the ....

D. Heckerman, J. S. Breese, and K. Rommelse. Decisiontheoretic troubleshooting. Communications of the ACM, 38(3):49-57, 1995.

....to nd a troubleshooting strategy s such that for all possible strategies s it holds that ECR(s ) ECR(s) 1. 3 Various setups of troubleshooting It is often reasonable to assume that if the device is malfunctioning, then there is exactly one fault in the device (single fault assumption) [3]. Otherwise the situation is referred to as multiple faults. When the faults are independent, we speak about independent faults [9] A solution of the troubleshooting task can be easily found in the case of independent actions when every action solves just one fault and all actions are pairwise ....

D. Heckerman, J. S. Breese, and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, 38(3):49-57, 1995.

....4 Models and Results We employed Bayesian model structure learning to infer predictive models from data and to identify key variables from the larger set of observations we collected. Over the last decade, there has been steady progress on methods for inferring Bayesian networks from data [6, 27, 12, 13]. Given a dataset, the methods typically perform heuristic search over a space of dependency models and employ a Bayesian score to identify models with the greatest ability to predict the data. The Bayesian score estimates p(modeljdata) by approximating p(datajmodel)p(model) Chickering et al. ....

D. Heckerman, J. Breese, and K. Rommelse. Decisiontheoretic troubleshooting. CACM, 38:3:49-57, 1995.

....over limited bandwidth channels. 11.1. Continual computation in diagnostic reasoning Over the last fifteen years, researchers have made significant advances on probabilistic representations and inference machinery that can support diagnosis and troubleshooting in challenging real world domains [8,32,34,36,66]. A significant number of interactive diagnostic systems based on probability and decision theory have been field in a variety of areas including healthcare, aerospace, and computing arenas. These systems, sometimes referred to as normative diagnostic systems, employ probabilistic models to ....

D. Heckerman, J. Breese, K. Rommelse, Decision-theoretic troubleshooting, Comm. ACM 38 (3) (1995) 49--57.

....a troubleshooting strategy s that minimizes the expected cost of repair ECR(s j ; among of all strategies s that have the same root as the AND OR graph corresponding to the troubleshooting task. It is often reasonable to assume that the problem has exactly one cause (single fault assumption) [2]. Otherwise the situation is referred as multiple faults. When the causes are independent we speak about independent faults [8] Solution of the troubleshooting task can be easily found in case of independent actions when every action solves just one cause and all actions are pairwise independent. ....

D. Heckerman, J. S. Breese, and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, 38(3):49-57, 1995.

....analyses. 4 Models and Results We employed Bayesian structure learning to infer predictive models from data and to identify key variables from the larger set of observations we collected. Over the last decade, there has been steady progress on methods for inferring Bayesian networks from data [6, 27, 12, 13]. Given a dataset, the methods typically perform heuristic search over a space of dependency models and employ a Bayesian score to identify models with the greatest ability to predict the data. The Bayesian score estimates p(modeljdata) by approximating p(datajmodel)p(model) Chickering, ....

D. Heckerman, J. Breese, and K. Rommelse. Decisiontheoretic troubleshooting. CACM, 38:3:49-57, 1995.

....analyses. 4 Models and Results We employed Bayesian structure learning to infer predictive models from data and to identify key variables from the larger set of observations we collected. Over the last decade, there has been steady progress on methods for inferring Bayesian networks from data [6, 27, 12, 13]. Given a dataset, the methods typically perform heuristic search over a space of dependency models and employ a Bayesian score to identify models with the greatest ability to predict the data. The Bayesian score estimates p(model data) by approximating p(data model)p(model) Chickering, ....

D. Heckerman, J. Breese, and K. Rommelse. Decisiontheoretic troubleshooting. CACM, 38:3:49--57, 1995.

....of Bayesian networks is that they provide a theoretical framework for combining statistical data with prior knowledge about the problem domain. Therefore, they are particularly useful in practical applications. Bayesian networks have been widely used for medical diagnosis[22] 9] troubleshooting[10], and in the communication network field, they have been proposed to diagnose faults in Linear Lightwave Networks[6] In [6] other methods have been used for detection and the Bayesian networks are used for diagnosis only. In this work we propose using a Bayesian network as a mechanism to combine ....

D. Heckerman, J.S. Breese, K. Rommelse, "Decision-theoretic troubleshooting," Communications of the ACM, vol. 38, March 1995, pp. 49-57.

....ecient decision theoretic troubleshooting of electro mechanical devices. In general, this task is NP complete, but under fairly strict assumptions, a greedy approach will yield an optimal sequence of actions, as discussed in the paper. This set of assumptions is weaker than the set proposed by Heckerman et al. 1995). However, the printing system domain, which motivated the research and which is described in detail in the paper, does not meet the requirements for the greedy approach, and a heuristic method is used. The method takes value of identi cation of the fault into account and it also performs a ....

....2 respect to likelihood (de Kleer Williams 1987) In decision theoretic troubleshooting costs and likelihoods are balanced in order to nd the next action. Decision theoretic troubleshooting was studied by Kalagnanam Henrion (1990) and it was extended to the context of Bayesian networks by Heckerman et al. 1995). They provide a framework for suggesting sequences of questions, repair actions, and con guration changes to obtain further information. By calculating a local eciency of the possible repair actions and continuously choosing the one of highest eciency, a repair sequence is established. Assuming ....

[Article contains additional citation context not shown here]

Heckerman, D., Breese, J. S. & Rommelse, K. (1995). Decision-theoretic troubleshooting, Communications of the ACM 38(3): 49-56. Special issue on real-world applications on Bayesian networks.

....knowledge about the problem domain. Therefore, they are particularly useful in practical applications. d(n) n p(n) W p[W,n p(n) p[W p(n) p[n p(n) Figure 1 Example of independence assumptions Bayesian networks have been widely used for medical diagnosis [15] 4] troubleshooting [5], and in the communication network field, they have been proposed to diagnose faults in Linear Lightwave Networks [6] In [6] other methods have been used for detection and the Bayesian networks are used for diagnosis only. In this work, we propose using a Bayesian network as a mechanism to ....

D. Heckerman, J.S. Breese, K. Rommelse, "DecisionTheoretic Troubleshooting," Communications of the ACM, vol. 38, March 1995, pp. 49-57.

....of a troubleshooting system for fixing faults in printing systems. The basic troubleshooting approach is outlined, concluding with a set of assumptions ensuring that a greedy approach will yield an optimal sequence of actions. The assumptions are weaker than the assumptions proposed by Heckerman, Breese Rommelse (1995). The printing system domain does not meet the requirements for the greedy approach, and a heuristic method is used. The method takes value of identification of the fault into account and it also performs a partial two step look ahead analysis. The validation process for the troubleshooter ....

....spends millions of dollars a year on customer support. Therefore, automating the troubleshooting process is highly beneficial for customer as well as supplier. Decision theoretic troubleshooting was studied by Kalagnanam Henrion (1990) and it was extended to the context of Bayesian networks by Heckerman et al. 1995). They provide a framework for suggesting sequences of questions, repair actions, and configuration changes to obtain further information. By calculating a local efficiency of the possible repair actions and continuously choosing the one of highest efficiency, a repair sequence is established. ....

[Article contains additional citation context not shown here]

Heckerman, D., Breese, J. S. & Rommelse, K. (1995). Decision-theoretic troubleshooting, Communications of the ACM 38(3): 49--57. Special issue on real-world applications on Bayesian networks.

....for Bayesian networks besides speech recognition. These include computer troubleshooting and medical diagnosis, among others. 7. 1 Computer Troubleshooting Perhaps the best known application of Bayesian networks in actual production is the printer problem troubleshooter for Microsoft Windows 95 (Heckerman et al. 1995). Figure 17 on page 20 gives this Bayesian network, as taken from the Microsoft Belief Networks (MSBN) program 3 . The network is used to help an amateur computer user to determine the source of the problem when he has trouble 3 http: www.research.microsoft.com msbn default.htm 20 IDIAP RR ....

Heckerman, D., Breese, J., and Rommelse, K. (1995). Decision-theoretic troubleshooting. Communications of the ACM , 38(3), 49-56.

....over N . D encodes the assumption that each variable x is independent of its nondescendants given its parents (x) This allows P to be expressed as P (N) Q x2N P (xj(x) A BN can be used to model our uncertain knowledge about a domain, e.g. medical diagnosis [5] equipment trouble shooting [6], financial forecasting [1] automated vehicles [3] etc. Figure 1 (a) shows a digital circuit and the DAG of a BN that models the circuit is shown in (b) An example conditional probability distribution associated with the variable f (output of a not gate) is given below: 3 i f e G5 G6 G2 ....

D. Heckerman, J.S. Breese, and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, pages 49--57, 1995.

....we address three practical modeling issues. 1 Introduction Bayesian networks (BNs) Pea88; Nea90; Jen96) provide a normative formalism for diagnosis based on probabilistic domain knowledge. In the past decade, researchers have studied how to model diagnostic problems using BNs (Hec90; DGH92; HBR95) and many algorithms have been proposed to perform inference in BNs (Pea88; Sha96; CGH97; Jen96) Most of these methods are based on a flat BN representation of the system to be diagnosed. As BNs become widely accepted, they are applied to larger and more complex problem domains. Construction of ....

D. Heckerman, J.S. Breese, and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, pages 49--57, 1995.

....suboptimal searching methods, such as n step lookahead, work well and give good results. Another issue which is hidden in the disscussion is the complexity for evaluating the conditional probabilities. In practice, using the method described in [16] the inference of probability is almost constant [17], even though the inference of an arbitrary belief network is NP hard [18] 4. A prototype on XUNET This section describes a prototype implementation of a fault management application on XUNET, one of the five Gigabit testbeds sponsored by the Corporation for National Research Initiatives. XUNET ....

D. Heckerman, J. S. Breese, and K. Rommelse, "Decision-theoretic troubleshooting," Comm. ACM, vol. 38, pp. 49--57, Mar. 1995.

....large Bayesian network model for printing system diagnosis. For decades, diagnosis has been a challenging application area for AI methodologies due to the complexity and data requirements involved. Much active research has been carried out in this area (de Kleer Williams 1987, Genesereth 1984, Heckerman et al. 1995, Breese Heckerman 1996) The purpose of diagnostic systems is to ultimately determine the set of faults that best explains a set of symptoms. A diagnostic system can request information from the world, and each time new information is obtained, it will update its current view of the world. The ....

....the set of faults that best explains a set of symptoms. A diagnostic system can request information from the world, and each time new information is obtained, it will update its current view of the world. The diagnostic engine in our work is based on the troubleshooting method suggested by Heckerman et al. 1995). This method provides suggestions for optimal observations, repairs, and configuration changes to obtain further information. The troubleshooter is myopic, i.e. it only has one step lookahead, and it is based on the single fault assumption. The diagnostic model for the printing system is being ....

Heckerman, D., Breese, J. S. & Rommelse, K. (1995). Decision-theoretic troubleshooting, Communications of the ACM 38(3): 49--56. Special issue on real-world applications on Bayesian networks.

....(for example the same component cannot be in more than one mode at the same time) This particular knowledge representation is suitable for diagnostic reasoning [6] Bayesian networks could be used instead of the probabilistic ATMS. Bayesian nets are used for diagnosis in [15] 14] and [5]. The advantage of the ATMS is that it can use focusing techniques. These techniques allow an ATMS to consider the components more likely to fail first. 4 Lifetimes: Why and How Intuitively, an older component is more likely to fail than a relatively new one 1 . Ignoring the notion of ....

J. Breese D. Heckerman and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, 38(3):49--57, 1995.

....1993] A good introduction to fuzzy systems is presented in [Negoita, 1984] The theoretical aspects of expert systems based on fuzzy logic are explored in [Gupta et al. 1985] which also brings several applications of these systems. 2.2. 3 Bayesian Networks Bayesian networks [Wright, 1921] apud [Heckerman et al. 1995b] constitute an interesting approach to the treatment of uncertainty. Through them, it is possible to produce inferences even when the available information is incomplete and inaccurate. Definition: A Bayesian network is a directed acyclic graph in which each node represents a random variable to ....

....under the aspects of performance and robustness, but it demands a great effort in the modeling of the object network, making it little recommendable for complex networks. Bayesian networks based methods were first utilized in 1921, in the analysis of harvesting results [Wright, 1921] apud [Heckerman et al. 1995b] and count on a solid mathematical basis. Each day, these methods win more acceptance, in the community of computing scientists, as a suitable option to the solution of problems involving uncertainty [Heckerman et al. 1995b] These factors contributed for the adoption of Bayesian networks in ....

[Article contains additional citation context not shown here]

David Heckerman, John S. Breese, and Koos Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, 38(3):49--57, Mar 1995.

....until the system is restored. 4 Conclusion This paper introduced a tractable algorithm for diagnosis repair planning where the agent goal is not to increase its knowledge, but rather to choose a sequence of actions that would restore the faulty system with minimal expected cost. Heckerman [ 5 ] presented a solution to the diagnosis repair problem assuming a single fault and each component is rst observed before repair. Srinivas [ 6 ] presented an algorithm that computes o line the optimal repair plan. He exploits the hierarchical description of the modeled system, assuming that each ....

....the hierarchical description of the modeled system, assuming that each component has only a small number of subcomponents to allow an exhaustive search. The algorithm presented here is a a straightforward extension of this two works. Namely, we do not assume a single fault thus generalizing [ 5 ] and we count for the possibility of sensing, that [ 6 ] does not. ....

Heckerman D., Breese J. and Rommelse K. Decisiontheoretic troubleshooting. In Communications of the ACM, 38(3):49-57. (1995)

....anomaly. In this subsection, we integrate fault diagnosis and the restoration planning in a single process. We show how to map a set of observation in the agent s domain to an optimal plan. The proposed approach is based on an approximate method for evaluating information gathering actions [9]. Let us assume that the agent has m possible sensors nodes S 1 ; S n that it can consult to reduce the uncertainty about which object is faulty. Furthermore let fs j 1 ; s j l j g be the set of possible values S j can take, and let ( denote the restoration plan initially ....

Heckerman D., Breese J. and Rommelse K. Decision-theoretic troubleshooting. In Communications of the ACM, 38(3), pp49-57, (1995)

....it has been recently transferred to a commercial system called INTELLIPATH, which is in used by several hundred medical and clinical sites. ffl Other Applications: Other application areas include, for example, software maintenance [5] natural language understanding [6] troubleshooting [11]. Despite these successful system deployments, systems designers who intend to use Bayesian networks like designers of knowledge based systems in general encounter the knowledge engineering bottleneck; that is, constructing a network manually is both time consuming and prone to error. ....

D. Heckerman, J. Breese, and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, 38(3):49--56, 1995.

....belief networks, probabilistic reasoning, probabilistic dependence models. 1 Introduction Belief networks (Bayesian networks, Markov networks and influence diagrams) 22, 20, 11] are becoming widely applied to AI tasks where representing and reasoning with uncertain knowledge are essential [2, 5, 7, 10, 19]. As an alternative and supplement to encoding probabilistic knowledge from domain experts, and driven by the data overload in the society, learning belief networks from data is being actively studied by many [9, 14, 24, 3, 28, 1, 4, 18, 8] The learned network is commonly used as a compact ....

D. Heckerman, J.S. Breese, and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, pages 49--57, 1995.

.... Other combination ideas are given by Ali Pazzani (1995) Ho (1995) and Shlien (1990) Chapter 8: Belief Networks Page 245: Applications of Bayesian belief networks are considered in in the March 1995 issue of Communications of the ACM by (Heckerman Wellman, 1995; Burnell Horvitz, 1995; Heckerman et al. 1995; Fung Del Favarro, 1995) Page 245, 258 262: Almond (1995) is a thesis length treatment of graphical computations for DempsterShafer belief functions, which have probability calculations as a special case. There are many possible join trees, and Almond considers other constructions which may ....

Heckerman, D., Breese, J. S. & Rommelse, K. (1995) Decision-theoretic troubleshooting.

....a posteriori explanations and maximum a posteriori expectation. Documentation, code and examples for JavaBayes can be downloaded from http: www.cs.cmu.edu fgcozman Research JavaBayes Home. As an example, consider a troubleshooting problem where the objective is to analyze the state of a car [20], which contains 17 variables and several deterministic and stochastic relationships. Suppose there is some imprecision in the probability values for two variables. First, take the variable BatteryAge, which has two values, Old and New. Suppose this variable is associated with an ffl contaminated ....

D. Heckerman, J. Breese, and K. Rommelse. Decision theoretic troubleshooting. Communications of the ACM, 38:49--57, 1995.

....in specific domains, probability theory, with its wellestablished mathematical foundation, provides sound and consistent knowledge representations for many domains [37, 44] 3. 2 Bayesian Networks One of the more popular knowledge representations accommodating uncertainty is the Bayesian Network [37, 7, 15, 26]. This knowledge representation models the probabilistic dependencies be3 R S Figure 3.1 Bayesian Network. The Bayesian Network corresponding to the rule: IF raining = true THEN sidewalk surface = wet with probability of 0.9 . Here R and S correspond to raining and sidewalk surface ....

Heckerman, David, et al. "Decision-Theoretic Troubleshooting," Communications of the ACM , 38 (3):49--57 (March 1995).

....apply. 1 Introduction Over the past few years, there has been a growing consensus within the AIcommunity that, as the real world is a noisy and nondeterministic place, it is often useful to model it as such. Modeling uncertainty has shown up in a variety of AI tasks as diverse as diagnosis (Heckerman et al. 1995), natural language processing (Charniak 1993) planning (Dean et al. 1993) and more. The different requirements of these tasks have resulted in the use of different stochastic modeling languages, such as Bayesian networks (Pearl 1988) and dynamic Bayesian networks (Dean and Kanazawa 1989) ....

D. Heckerman, J. Breese, and K. Rommelse. Decisiontheoretic troubleshooting. CACM, 38(3):49--57, 1995.

....section, we describe a set of assumptions under which it is possible to identify an optimal sequence of observations and repair actions in time proportional to the number of components in the device, without explicitly constructing and rolling back a decision tree. The approach is described in Heckerman et al. 1995). Let us suppose that the device has n components c 1 ; c n and each component is in exactly one of a finite set of states. We assume 1 1 The appropriateness of these assumptions is discussed in Heckerman et al. 1995) 2 1. There is only one problem defining variable in the Bayesian ....

....and rolling back a decision tree. The approach is described in Heckerman et al. 1995) Let us suppose that the device has n components c 1 ; c n and each component is in exactly one of a finite set of states. We assume 1 1 The appropriateness of these assumptions is discussed in Heckerman et al. 1995). 2 1. There is only one problem defining variable in the Bayesian network for the device. This variable represents the functional status of the device. One of the states of this variable must correspond to normal operation. In Figure 1, the node labeled Printer Output is the problem defining ....

[Article contains additional citation context not shown here]

Heckerman, D., Breese, J., and Rommelse, K. (1995). Decision-theoretic troubleshooting. Communications of the ACM, 38:49--57.

....providing the model. In an application involving the effect of drugs on white blood cell counts, we found the temporal version of causal independence to be a more natural method for interacting with the expert [Heckerman, 1993] In contrast, in a number of hardware troubleshooting applications [Heckerman et al. 1995a] we found the amechanistic form to be more effective. 4 Inference Improvement in Bayesian Networks In the previous section, we saw that there are two potential sources for gains in inference efficiency: 1) reduction in the size of parent sets afforded by (singly) decomposable causal ....

Heckerman, D., Breese, J., and Rommelse, K. (1995a). Decision-theoretic troubleshooting. Communications of the ACM, 38:49--57.

No context found.

Heckerman, D., J. S. Breese, and Koos Rommelse, "Decision Theoretic Troubleshooting," Communications of the ACM, Vol. 38, No. 3, pp. 49-56, March 1995.

No context found.

D. Heckerman, J. Breese, and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, 38:49--57, 1995.

No context found.

D. Heckerman, J.S. Breese, K. Rommelse. Decision-Theoretic Troubleshooting. Communications of the ACM, 38(3):49-57, March 1995.

No context found.

D. Heckerman, J. S. Breese, and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, 38(3):49--57, 1995.

No context found.

D. Heckerman, J.S. Breese, and K. Rommelse. Decision-theoretic troubleshooting. Communications of the ACM, pages 49--57, 1995.

No context found.

HECKERMAN,D.E.,BREESE, J., AND ROMMELSE,K. 1995. Decision-theoretic troubleshooting.

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