A.H. Doan and P. Haddawy. Decision-theoretic refinement planning: Principles and application. Technical Report TR-95-01-01, Dept. of Elect. Eng. & Computer Science, University of Wisconsin-Milwaukee, January 1995. Available via anonymous FTP from pub/techreports at ftp.cs.uwm.edu.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....to reduce the set of actions, and abstracting groups of branches within an action. In this paper we show how abstract macro operators can be generated which compactly represent a sequence of actions and thus reduce the cost of projection. A comprehensive theory of abstraction can be found in [ 2 ] . Our work on generating macro operators was motivated by our application of the drips decisiontheoretic refinement planning system to the problem of selecting the optimal test treat strategy in a particular medical domain [ 5 ] drips efficiently searches the space of possible plans to identify ....

....deal with mass assignments where we cannot determine in advance which state belongs to a focal elemment B, because B is represented by a set of constraints. In such an environment, general mass assignments have more expressive power than simple mass assignments. For a more detailed discussion see [ 2 ] . Throughout the paper we will be talking about action descriptions, but for brevity will refer to them simply as actions. Actions serve as transformations from a set of probability distributions pre into a set of probability distributions post . The set post is a function of a and pre , ....

[Article contains additional citation context not shown here]

A.H. Doan and P. Haddawy. Decision-theoretic refinement planning: Principles and application. Technical Report TR-95-01-01, Dept. of Elect. Eng. & Computer Science, University of Wisconsin-Milwaukee, January 1995. Available via anonymous FTP from pub/techreports at ftp.cs.uwm.edu.

....expected utility of the plan can be computed in a manner similar to computing the expected utility of a concrete plan, and the interval of the expected utility of a plan includes the expected utility of any of its subplans. We have reported abstraction techniques satisfying these requirements in [6]. Consider the case when the lower bound of plan p 1 s utility interval is greater than the upper bound of plan p 2 s utility interval, from the abstraction requirements it follows that any subplan of p 2 has lower expected utility than any subplan of p 1 , and therefore cannot be the optimal ....

A.H. Doan and P. Haddawy. Decisiontheoretic refinement planning: Principles and application. Technical Report TR-95-01-01, Dept. of Elect. Eng. & Computer Science, University of Wisconsin-Milwaukee, January 1995. Available via anonymous FTP from pub/techreports at ftp.cs.uwm.edu.

....on the paired branches; and the effect is the composition of the effects. We have implemented tools that automatically create inter action abstractions [ 5 ] and sequential abstractions [ 3 ] For a general theory of action abstraction which includes intra action and sequential abstraction see [ 4 ] . 3.2 The drips Planner A planning problem is described in terms of an initial state distribution, a set of action descriptions, and a utility function. The space of possible plans is described by an abstraction decomposition network, supplied by the user. An abstract action has one or more ....

A.H. Doan and P. Haddawy. Decision-theoretic refinement planning: Principles and application. Technical Report TR-95-01-01, Dept. of Elect. Eng. & Computer Science, University of Wisconsin-Milwaukee, January 1995. Available via anonymous FTP from pub/techreports at ftp.cs.uwm.edu.

.....3 present=present 30 ton deli = ton fuel=fuel 6 ton=0 fuel 7 no effect Mpost .8 .7 ton deli[0] 0 present=10 fuel[0] 8, 10] ton[0] 10, 20] ton deli[10] 8, 16] fuel[10] 3, 5] ton[10] 0 .8 . 3 ton deli[0] 0 present=30 fuel[0] 8, 10] ton[0] 10, 20] ton deli[30] 10, 20] fuel[30] [2, 4] ton[30] 0 .2 ton deli[0] 0 present=0 fuel[0] 5, 6] ton[0] 10, 20] project( Deliver , Mpre ) Figure 1: Initial state represented as a general mass assignment, action description, and the post projection state. 2: for P pre 2 X; exec(a; P pre ) fP post jP post (A) P b P pre (b) Delta P ....

....states the correctness of this projection rule. Theorem 1 Given P pre 2 (M pre ) action a, and P post 2 exec(a; P pre ) there exists m post 2 M post , where M post is calculated using (4) such that P post 2 (m post ) Consequently exec(a; M pre ) M post ) For a proof of this theorem, see [4]. It is not hard to prove that Theorem 1 is true even when E ij (B) is defined such that E ij (B) E ij (b) for only those states b such that b 2 B c i . Projection rule 1 states that in order to compute M post , we project each branch (triple) of a on each branch of M pre , where for ....

[Article contains additional citation context not shown here]

A.H. Doan and P. Haddawy. Decision-theoretic refinement planning: Principles and application. Technical Report TR-95-01-01, Dept. of Elect. Eng. & Computer Science, University of Wisconsin-Milwaukee, January 1995. Available via anonymous FTP from pub/techreports at ftp.cs.uwm.edu.

No context found.

Doan, A. H., and Haddawy, P. 1995b. Decision-theoretic refinement planning: Principles and application. Technical Report TR-95-01-01, Dept of Elect. Eng & Computer Science, University of Wisonsin-Milwaukee.

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