A. Doan and P. Haddawy. Sound abstraction of probabilistic actions in the constraint mass assignment framework. In E. Horvitz and F. V. Jensen, editors, Proceedings of the Twelfth Annual Conference on Uncertainty in Artificial Intelligence (UAI-96), pages 228-235, San Francisco, California, 1996. Morgan-Kaufmann.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....policies that would be optimal if the imprecision in the model was resolved in the best and worst possible way, respectively. This is essentially a variation of max min and maxmax rules. They do not consider possibility of indeci sion. Another related work is the work of Haddawy s group, e.g. [3, 5]. Their motivation is mainly abstraction for efficient solution of planning problems. They consider more general setting for sequential decision making (they do not assume the Markov property) but restrict the uncertainty representation to lower upper probabilities on a subclass of 7 (S) in my ....

A. Doan and P. Haddawy. Sound abstraction of probabilistic actions in the constraint mass assignment framework. In E. Horvitz and F. V. Jensen, editors, Proceedings of the Twelfth Annual Conference on Uncertainty in Artificial Intelligence (UAI-96), pages 228-235, San Francisco, California, 1996. Morgan-Kaufmann.

....into account the current robot location. We want to improve our method of estimating the expected cost in the more abstract levels. Our present estimation is ad hoc and we have not calculated how suboptimal it is. It would be interesting to consider sound abstraction like that of Doan and Haddawy [4]. There the approximation of the expected cost of abstract nodes is specified using upper and lower bounds on the expected cost so it is more accurate. As mentioned in the introduction, the use of abstract representations to obtain approximate solutions at a lower computational cost has been ....

A. Doan and P. Haddawy. Sound abstraction of probabilistic actions in the constraint mass assignment framework, in: Proceedings of UAI'96, 228-235, 1996.

....account the current robot location. We want to improve our method of estimating the expected cost in the more abstract levels. Our present estimation is ad hoc and we have not calculated how suboptimal it is. It would be interesting to consider sound abstraction like that of Doan and Haddawy [ Doan and Haddawy, 1996 ] There the approximation of the expected cost of abstract nodes is specified using upper and lower bounds on the expected cost so it is more accurate. a b c d a b c d 0 10 20 30 40 50 60 50 0 50 100 150 200 250 Room Revised hierarchical (no sensors) 0 10 20 30 40 50 60 10 0 ....

A. Doan and P. Haddawy. Sound abstraction of probabilistic actions in the constraint mass assignment framework. In UAI'96, pages 228--235, 1996.

....into account the current robot location. We want to improve our method of estimating the expected cost in the more abstract levels. Our present estimation is ad hoc and we have not calculated how suboptimal it is. It would be interesting to consider sound abstraction like that of Doan and Haddawy [4]. There the approximation of the expected cost of abstract nodes is specified using upper and lower bounds on the expected cost so it is more accurate. As mentioned in the introduction, the use of abstract representations to obtain approximate solutions at a lower computational cost has been ....

A. Doan and P. Haddawy. Sound abstraction of probabilistic actions in the constraint mass assignment framework. In UAI'96, pages 228--235, 1996.

...., which is defined as: w final = fpl(w ini )jpl 2 plg. The expected utility of a plan pl is defined as the u(pl(w ini ) fu(pl(w ini ) jpl 2 plg. These notions of abstract actions and plans are introduced in [7] to model sound abstraction, the central concept of the work of Haddawy, et.al. [8, 3] on abstraction based probabilistic planning. In this work, an abstract action A is called a sound abstraction of n concrete actions A 1 ; A n if for any initial world w ini 2 S, we have A i (w ini ) 2 A(w ini ) i = 1; n. For example, the abstract action A in Figure 1.b is a ....

A. Doan and P. Haddawy. Sound abstraction of probabilistic actions in the constraint mass assignment framework. In Proceedings of the Twelfth Conference on Uncertainty in Artificial Intelligence, pages 228-- 235, Portland, August 1996.

....of actions and to reduce the cost of plan evaluation. Introduction Decision theoretic approaches to planning, while representationally appealing, tend to be extremely computationally costly. To cope with this complexity, researchers have employed abstraction techniques (Boutilier Dearden 1994; Doan Haddawy 1996). Haddawy et al. (Haddawy Suwandi 1994; Doan Haddawy 1996) identify three types of abstraction that can be used to reduce the complexity of decisiontheoretic planning when using a tree based representation. While the abstraction techniques have been quite successful in reducing the complexity ....

....Decision theoretic approaches to planning, while representationally appealing, tend to be extremely computationally costly. To cope with this complexity, researchers have employed abstraction techniques (Boutilier Dearden 1994; Doan Haddawy 1996) Haddawy et al. (Haddawy Suwandi 1994; Doan Haddawy 1996) identify three types of abstraction that can be used to reduce the complexity of decisiontheoretic planning when using a tree based representation. While the abstraction techniques have been quite successful in reducing the complexity of planning, the tree representation provides little help in ....

[Article contains additional citation context not shown here]

Doan, A., and Haddawy, P. 1996. Sound abstraction of probabilistic actions in the constraint mass assignment framework. In Proceedings of the Twelfth Conference on Uncertainty in Artificial Intelligence, 228--235.

....and is a generalization of mass assignment representations. Third, we present sound abstraction procedures for three types of action abstraction. Finally, we end the paper with a discussion of related work on sound and approximate abstraction. Due to space limitation, the reader is referred to [ 5 ] for a more detailed discussion of the topics of this paper, as well as discussions of applications, estimating loss due to abstraction, and automatically constructing abstraction hierarchies. 2 SOUND PROBABILISTIC ACTION ABSTRACTION In this section we present a general representation framework ....

....procedures for the mass assignment framework. To the best of our knowledge, most existing work on developing this framework presents projection rules that iterate through all subsets of the state space and so cannot be applied to domains with large state space. For a discussion on this topic see [ 4, 5 ] . We have found that a generalized representation of the mass assignment framework, which we call the constraint mass assignment framework, is well suited for practical projection rules and sound abstraction. 0.8 0.2 fuel = 4, 8] fuel = 2, 6] 0.8 0.2 fuel = 6 fuel = 3 [0.8, 1] 0, 0.2] fuel ....

[Article contains additional citation context not shown here]

A. Doan and P. Haddawy. Sound abstraction of probabilistic actions in the constraint mass assignment framework. Technical report, University of WisconsinMilwaukee, March 1996. Available via anonymous FTP from pub/techreports at ftp.cs.uwm.edu.

....the process of evaluating and eliminating suboptimal plans depends mainly on three factors: i) the number of the branches of the actions in a plan, ii) the number of the actions in a plan, and (iii) the number of alternative plans in the plan space. Three abstraction techniques are discussed in [6] to reduce the complexity of the planning process. The intra action abstraction technique, initially 7 The convex hull of a set of real numbers is the smallest interval that contains all the numbers. 8 The problem can be interpreted as the problem of packing objects of different values into a ....

A. Doan and P. Haddawy. Sound abstractions of probabilistic actions in the constraint mass assignment framework. In Proceeding of the Twelfth Conference on Uncertainty in Artificial Intelligence, 1996.

No context found.

A. Doan and P. Haddawy, "Sound Abstraction of Probabilistic Actions in the Constraint Mass Assignment Framework ", Proc. UAI'96, pp. 228--235. D. Draper and S. Hanks, "Localized Partial Evaluation of Belief Networks", Proc. UAI'94. D. Dubois and H. Prade, Fuzzy Sets and Systems: Theory and Applications, Academic Press, N.Y., London, 1980.

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