D. Heckerman, E. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE Trans. Pattern Analysis and Machine Intelligence, 15:292--298, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....applications a non myopic approach in which nodes are selected groupwise outperforms any method based on a myopic approach. Naively adopting a non myopic approach, however, poses serious computational problems. Further research aimed at gaining insight in solving these problems is underway [ Heckerman et al. 1993 ] Since adopting a myopic approach to evidence gathering is quite common in diagnostic knowledge based systems, we will equally take this approach in this paper. Secondly, the number of nodes in the belief network representing disorders is often restricted to one. This restriction prohibits ....

D.E. Heckerman, E.J. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 15, pp. 292--298, 1993.

....since then all possible combinations of values for all hypothesis vertices have to be considered. We will here equally take up the two assumptions mentioned above. We would like to note, however, that recent research results indicate that the simplifying assumptions may be eased to some extent [Heckerman et al. 1993, van der Gaag Wessels, 1994] Selective evidence gathering for diagnostic problem solving with a belief network may now be envisioned as outlined below in pseudo code. The evidence gathering procedure takes the digraph G of a belief network and the set E of all yet uninstantiated evidence ....

D.E. Heckerman, E.J. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 15, 1993, pp. 292 - 298.

....in an interactive mode. Specifically, given a utility model, we can use value of information analyses to guide selection of additional information and refine a disease hypothesis. We are developing approximate methods to compute non myopically the value of new information given an evidence set [18]. Techniques such as these may allow hypotheses to be iteratively generated and refined in a manner similar to the intended use of QMR [4] 4 . 8 Shortcomings of the QMR DT Belief Network Model Because the current QMR DT KB is a straightforward reformulation of the INTERNIST 1 KB, both KBs ....

Heckerman DE, Horvitz EJ, Middleton B. An Approximate Nonmyopic computation for Value of Information. Knowledge Systems Laboratory Memo no. KSL-91-15. Stanford CA: Stanford University, 1991. Middleton B, Shwe M, Heckerman D,et al. 32

....since then all possible combinations of values for all hypothesis vertices have to be considered. We will here equally take up the two assumptions mentioned above. We would like to note, however, that recent research results indicate that the simplifying assumptions may be eased to some extent [Heckerman et al. 1993, van der Gaag Wessels, 1994] Selective evidence gathering for diagnostic problem solving with a belief network may now be envisioned as outlined below in pseudo code. The evidence gathering procedure takes the digraph G of a belief network and the set E of all yet uninstantiated evidence ....

D.E. Heckerman, E.J. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 15, 1993, pp. 292 - 298.

....but vision problems are usually too complex for this approach. Another alternative is to perform some kind of heuristic search of the space of all possible sequences. A search method that uses small subsequences with good evaluations to find larger sequences with good evaluations is presented in [Heckerman et al. 1993]. An Approximate Solution Based on Utility Information An approximate solution can be based on utility information. The idea here is to use the utility information to compute the value of individual action choices, or perhaps short (2 3) subsequences of action choices, and then to glue the ....

D. Heckerman, E. Horvitz, and B. Middleton, "An Approximate Nonmyopic Computation for Value of Information," IEEE Transactions on Pattern Analysis and Machine Intelligence, 15(3):292--298, 1993.

....the NPhardness of exact solution by proving NP hardness of a special case called fault trees. However, Coo87] is the better known result. 2.6 Other forms of queries 2.6. 1 Value of Information How valuable would one additional piece of evidence be: How66] DBL90] Mat90] ZQP93a] HHM91] HHM93] JL94] Eza94] 2.6.2 Fast Retraction of Evidence How does result change if one (or more) item(s) of evidence is not included (related also to Sensitivity Analysis) Daw92] Jen95] 2.6.3 Sensitivity Analysis How sensitive are answers to model probabilities: Kor90] HS93] Las93] ....

David Heckerman, Eric Horvitz, and Blackford Middleton. An approximate nonmyopic computation for value of information. IEEE Transactions on Pattern Analysis and Machine Intelligence, 15(3):292--298, March 1993.

....proved the NPhardness of exact solution by proving NP hardness of a special case called fault trees. However, Coo87] is the better known result. 2.6 Other forms of queries 2.6. 1 Value of Information How valuable would one additional piece of evidence be: How66] DBL90] Mat90] ZQP93a] HHM91] HHM93] JL94] Eza94] 2.6.2 Fast Retraction of Evidence How does result change if one (or more) item(s) of evidence is not included (related also to Sensitivity Analysis) Daw92] Jen95] 2.6.3 Sensitivity Analysis How sensitive are answers to model probabilities: Kor90] HS93] ....

David Heckerman, Eric Horvitz, and Blackford Middleton. An approximate nonmyopic computation for value of information. In Uncertainty in Artificial Intelligence: Proceedings of the Seventh Conference, pages 135--141. Morgan Kaufmann, 1991.

....information sources which maximize the metric. We consider both these issues and present workable approaches. Many authors have recognized the importance of test selection in larger applications. Recent proposed approaches for probabilistic belief networks include: Jensen and Liang (1994) Heckerman, et al. 1993), and Almond (1993) In this paper, we propose a semiautomated myopic strategy which synthesizes the critiquing approach of Miller (1983) and Good s idea of a quasi utility (Good and Card, 1971) We develop search strategies based on these ideas and demonstrate them in the context of a simple ....

....1971) We develop search strategies based on these ideas and demonstrate them in the context of a simple imaging application. We also address a limited form of the nonmyopic test selection problem, and propose and demonstrate a Markov chain Monte Carlo solution that generalizes the recent work of Heckerman, et al. 1993). Section 9.2 reviews basic test selection techniques without particular reference to belief 1 Learning from Data: AI and Statistics V. Edited by D. Fisher and H. J. Lenz. c fl1996 Springer Verlag. 2 Address for correspondence: Department of Statistics, GN 22, University of Washington, Seattle, ....

[Article contains additional citation context not shown here]

Heckerman, D., E. Horvitz, and B. Middleton[1993]. "An Approximate Nonmyopic Computation for Value of Information." IEEE Transaction of Pattern Analysis and Machine Intelligence (15), 292--298.

....approach in which variables are selected groupwise might outperform any method for selective evidence gathering based on a myopic approach. Adopting a non myopic approach, however, poses some serious computational problems. Further research is aimed at gaining insight in solving these problems, [Heckerman et al. 1993]. Now envision performing the task of selective evidence gathering in the context of a diagnostic belief network. It may be outlined as follows: 1. Select the variable that is expected to contribute most to the confirmation or disconfirmation of the hypothesis; 2. Request the value of the selected ....

D. Heckerman, E. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 15, pp. 292--298, 1993.

....applications a non myopic approach in which nodes are selected groupwise outperforms any method based on a myopic approach. Naively adopting a non myopic approach, however, poses serious computational problems. Further research aimed at gaining insight in solving these problems is underway [ Heckerman et al. 1993 ] Since adopting a myopic approach to evidence gathering is quite common in diagnostic knowledge based systems, we will equally take this approach in this paper. Secondly, the number of nodes in the belief network representing disorders is often restricted to one. This restriction prohibits ....

D.E. Heckerman, E.J. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 15, pp. 292--298, 1993.

....assumptions (or constraints) i) no competition, and (ii) one step horizon. In other words, given a set of information sources (e.g. sets of experts or tests) the best source to poll is the one that would be the most valuable if action were taken immediately after its information were provided [58]. In order to avoid the types of constraints imposed by a myopic policy, several researchers have proposed alternative, non myopic, approximation methods [58] The procedure for determining the value of missing information is an example of a sensitivity analysis. In this instance, what is being ....

.... the best source to poll is the one that would be the most valuable if action were taken immediately after its information were provided [58] In order to avoid the types of constraints imposed by a myopic policy, several researchers have proposed alternative, non myopic, approximation methods [58]. The procedure for determining the value of missing information is an example of a sensitivity analysis. In this instance, what is being analyzed is the sensitivity of the decision to the addition of information; standard decision analytic sensitivity analyses have long addressed this issue. ....

D. Heckerman, E. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. In P. Smets B. D. D'Ambrosio and P. P. Bonissone, editors, Proceedings of the Seventh Conference on Uncertainty in Artificial Intelligence, pages 135--141, University of California, Los Angeles, July 1991.

No context found.

D. Heckerman, E. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE Trans. Pattern Analysis and Machine Intelligence, 15:292--298, 1993.

No context found.

D. Heckerman, E. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE TPAMI, 15:292--298, 1993.

No context found.

D. Heckerman, E. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE Trans. Pattern Anal. Mach. Intell., 15(3):292--298, 1993.

No context found.

D. Heckerman, E. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE Trans. Pattern Analysis and Machine Intelligence, 15:292--298, 1993.

No context found.

D. E. Heckerman, E. J. Horvitz, and B. Middleton. An approximate nonmyopic computation for value of information. IEEE Transactions on Pattern Analysis and Machine Intelligence, 15:292--298, 1993.

No context found.

D. Heckerman, E. Horvitz, and B. Middleton, "An approximate nonmyopic computation for value of information, " IEEE Trans. on PAMI, 15(3), pp.292-298, 1993.

No context found.

Heckerman D, Horvitz E, Middleton B. An approximate nonmyopic computation for value of information. Proceedings of the Seventh Conference on Uncertainty in Artificial Intelligence, 1991: 135-141. 41

No context found.

D. Heckerman, E. Horvitz and B. Middleton, "An Approximate Nonmyopic Computation for Value of Information", IEEE Transaction of Pattern Analysis and Machine Intelligence, 1993.

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