## Bayes Network "Smart" Diagnostics (2004)

Citations: | 1 - 0 self |

### BibTeX

@MISC{Agosta04bayesnetwork,

author = {John Mark Agosta and Corporate Technology Group and Intel Corporation},

title = {Bayes Network "Smart" Diagnostics},

year = {2004}

}

### OpenURL

### Abstract

Formal diagnostic methods are emerging from the machine-learning research community and beginning to find application in Intel. In this paper we give an overview of these methods and the potential they show for improving diagnostic procedures in operational environments. We present an historical overview of Bayes networks and discuss how they can be applied to diagnosis. We then give an illustration of how they can model the faults in a vacuum subsystem of a manufacturing tool.

### Citations

7341 |
Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference
- Pearl
- 1988
(Show Context)
Citation Context ...tive reasoning methods stagnated when purely symbolic approaches became popular in the early 1980s. The current blossoming of Bayesian methods coincides with Pearl’s introduction of “Belief networks” =-=[20]-=- and Heckerman and others work on the Pathfinder project [9], as an improvement on the Mycin medical diagnostic rule-based expert system. Bayes Network “Smart” Diagnostics 363sIntel Technology Journal... |

1324 |
Local Computations with Probabilities on Graphical Structures and Their Application to Expert Systems
- Lauritzen, Spiegelhalter
- 1988
(Show Context)
Citation Context ... that is more widely applicable than conventional statistics-based derivations, but maintains consistency with them. The conclusions presented here are based on an argument first presented by R.T.Cox =-=[7, 13, 18]-=- He begins with three premises: I. The uncertainty of an event is described by a real number. It is possible to find an equivalence between any real, uncertain value and a certain equivalent. For inst... |

652 | Probabilistic Networks and Expert Systems - Cowell, Dawid, et al. - 1999 |

647 | Learning in Graphical Models
- Jordan, editor
- 1998
(Show Context)
Citation Context ...rn Bayes network probabilities. Bayesian statistics consider the proper way to combine expert judgment and statistical knowledge. For a comprehensive review see especially the article by Heckerman in =-=[16]-=-. The models we develop work fine with the rank-ordered probability numbers estimated by the technicians. The arguments why Bayes net methods work well with order of magnitude probabilities and other ... |

615 | Bayesian Networks and Decision Graphs - Jensen - 2001 |

145 |
Information Value Theory
- Howard
- 1966
(Show Context)
Citation Context ...lue of information.” Strictly, value of information refers to the increase in expected value of the outcomes under the optimal decision were the information value available when the decision was made =-=[12]-=-. In our example, if we set diagnostic evidence so that the “rate-of-rise” test is set to “fail to pump down” we get the recalculated ranked probabilities shown in Figure 10. Note that the vacuum pump... |

117 |
The Algebra of Probable Inference
- Cox
- 1961
(Show Context)
Citation Context ... that is more widely applicable than conventional statistics-based derivations, but maintains consistency with them. The conclusions presented here are based on an argument first presented by R.T.Cox =-=[7, 13, 18]-=- He begins with three premises: I. The uncertainty of an event is described by a real number. It is possible to find an equivalence between any real, uncertain value and a certain equivalent. For inst... |

58 | The sensitivity of belief networks to imprecise probabilities: an experimental investigation
- Pradhan, Henrion, et al.
- 1996
(Show Context)
Citation Context ...obability numbers estimated by the technicians. The arguments why Bayes net methods work well with order of magnitude probabilities and other methods don’t are discussed extensively in the literature =-=[21]-=-. History of Bayes Network Diagnostics Early work in perceptrons, precursors to neural networks, was sympathetic to Bayesian approaches. Progress in Bayesian approaches to automated reasoning along wi... |

46 | Probability Theory: The Logic of
- Jaynes
- 2003
(Show Context)
Citation Context ...which specific cause or causes can explain a breakdown. Why not use an entirely qualitative (e.g., symbolic) calculus? Because there aren’t any that don’t reduce to computing with probability numbers =-=[14]-=-, and because there is a bonus in using probability-based logic–it can be extended to statistical techniques. This has a direct benefit when data are available from which to learn Bayes network probab... |

31 | A Comparison of decision analysis and expert rules for sequential analysis - Kalagnanam, Henrion - 1990 |

30 | Building probabilistic networks: where do the numbers come from - Druzdzel, Gaag - 2000 |

23 | Decision-theoretic troubleshooting: A framework for repair and experiment
- Breese, Heckerman
- 1996
(Show Context)
Citation Context ...ifferent costs to performing tests. The full decision optimization problem can be addressed by embedding the diagnostic Bayes network in an optimization framework, but this is not typically justified =-=[10]-=-. Interestingly, see BayerZubek [5] for a novel Bayesian approach that does not take advantage of a Bayes network, but does compute an optimal diagnostic policy. Cox’s Theorem as the Foundation for Pr... |

22 | Selective evidence gathering for diagnostic belief networks
- Gaag, Wessels
- 1993
(Show Context)
Citation Context ...of Diagnostic Observations: Finding the Best Test Sequence The challenge in diagnostic reasoning is computing which is the best observation to select next, to make progress towards finding the faults =-=[23]-=-. This is called differential diagnosis, at the stage in the diagnosis when the current information suggests several candidate faults with a high probability, to be confirmed or refuted to find what t... |

17 |
Learning diagnostic policies from examples by systematic search
- Bayer-Zubek
- 2004
(Show Context)
Citation Context ...The full decision optimization problem can be addressed by embedding the diagnostic Bayes network in an optimization framework, but this is not typically justified [10]. Interestingly, see BayerZubek =-=[5]-=- for a novel Bayesian approach that does not take advantage of a Bayes network, but does compute an optimal diagnostic policy. Cox’s Theorem as the Foundation for Probabilistic Logic Probability is a ... |

13 |
Probability via Expectation
- Whittle
- 1992
(Show Context)
Citation Context ... 2004 II. the event has value 1 and its negation, the absence of the event, has value 0. In most expositions the concept of expectation is derived from probability. In this case, similarly to Whittle =-=[24]-=- the reverse is more natural. Probabilities combine in common sense ways. When events are certain, probabilistic logic is reduced to familiar Aristotelian logic. When probabilities can be defined by f... |

5 |
Decision theoretic troubleshooting
- Heckerman, Breese, et al.
- 1995
(Show Context)
Citation Context ...ons approaches zero. This approximate guided search method is called myopic test selection. Simulations in typical diagnostic models have shown that myopic test selection closely approximates optimal =-=[11]-=-. The mutual information computation over all possible observations as required by myopic test selection may be onerous, and further approximations may be necessary [13]. Continuing with our example, ... |

4 |
An evaluation of the diagnostic accuracy of Pathfinder
- Heckerman, Nathwani
- 1992
(Show Context)
Citation Context ...hes became popular in the early 1980s. The current blossoming of Bayesian methods coincides with Pearl’s introduction of “Belief networks” [20] and Heckerman and others work on the Pathfinder project =-=[9]-=-, as an improvement on the Mycin medical diagnostic rule-based expert system. Bayes Network “Smart” Diagnostics 363sIntel Technology Journal, Volume 8, Issue 4, 2004 Figure 1: Chamber, valve, and pump... |

3 |
Active Fusion for Diagnosis Guided by Mutual Information Measures”, Proceeding of the 2 nd International Conference on Information Fusion
- Agosta, Weiss
(Show Context)
Citation Context ...ts of driving the probability of the faults to extremes–the probability of true faults to one and the rest to zero. An appropriate way to measure this is to compute the entropy over the set of faults =-=[3]-=-. Thus the most efficient set of observations is the set that maximizes the decrease in entropy of the fault set. Since the Bayes network can be used to compute the full joint probability distribution... |

3 |
Support for Multiple Cause Diagnosis with Bayesian Networks
- Jagt
- 2002
(Show Context)
Citation Context ... that is more widely applicable than conventional statistics-based derivations, but maintains consistency with them. The conclusions presented here are based on an argument first presented by R.T.Cox =-=[7, 13, 18]-=- He begins with three premises: I. The uncertainty of an event is described by a real number. It is possible to find an equivalence between any real, uncertain value and a certain equivalent. For inst... |

1 |
The Fastlane to Auto Parts and Accessories http://www.partsamerica.com
- America
(Show Context)
Citation Context ...cs are used in several commercial implementations, most notably by Microsoft, in General Electric’s Condition Forecaster * tool and for troubleshooting websites (see, for example, “Parts America.com” =-=[19]-=-). Intel has sponsored an annual Bayesian Application Workshop in cooperation with the Uncertainty in Artificial Intelligence conference where commercial implementation projects are presented [4]. PRO... |