Othar Hansson and Andrew Mayer. The optimality of sat- isficing solutions. In Proceedings of the Fourth Workshop on Uncertainty in Artificial Intelligence, 1988.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....problems that arise in controlling robots. 1.3. 4 Applications We have already mentioned several application areas in which the decision theoretic control of reasoning has been explored, including planning [Elkan, 1990, Drummond and Bresina, 1990] game tree search [lussell and Wefald, 1989a, Hansson and Mayer, 1988] medical decision making [Horvitz, 1988] and robot control [Dean and Boddy, 1988, Boddy and Dean, 1989] In addition, decision theoretic deliberation scheduling frameworks have been proposed for computer vision [Levitt et al. 1988] industrial control [Agogino et al. 1988, lamamurthi and ....

....expands another node or nodes in the search tree, possibly causing a change in the ranking of possible next moves or in the best known solution. Using decision analytic methods to determine how (and whether) to expand search trees is currently an active research area [lussell and Wefald, 1989b, Hansson and Mayer, 1988]. Probabilistic Inference In the most common form of probabilistic inference, we start with a set of random variables, information on how the values of these variables depend on one another, and a set of observations fixing values for some of the variables. We then calculate the posterior ....

Othar Hansson and Andrew Mayer. The optimality of sat- isficing solutions. In Proceedings of the Fourth Workshop on Uncertainty in Artificial Intelligence, 1988.

....the presence of unexpected interruption, and simplifying the problem of optimal or near optimal deliberation scheduling. Both of these approaches are discussed in more detail in Section 4. The use of deliberation scheduling has been explored in several domains, including planning [13, 11] search [35, 19, 28], medical decision making [26] and robot control [10, 5] 3 Decision Theory and the Control of Inference Probability and decision theory are methods for dealing with uncertain information and outcomes on the basis of a small set of axioms concerning rational behavior [31] If you accept the ....

Hansson, Othar and Mayer, Andrew, The Optimality of Satisficing Solutions, Proceedings of the Fourth Workshop on Uncertainty in Artificial Intelligence, Minneapolis, MN, 1988.

.... while resources (time and memory) are limited has led some researchers to consider something like a comprehensive value of computation, explicitely taking the quality of the solution into account as well as the time (and other resources such as memory) which have been required to derive it [8, 12]. Corresponding methods fall back on principles from probability theory, utility theory, and (statistical) decision theory. The latter is a well established approach in the field of economics but it is relatively new for AI. So far, probabilistic inference and ideas from decision theory have been ....

O. Hansson and A. Mayer. The optimality of satisficing solutions. In Proceedings Fourth Workshop on Uncertainty in Artificial Intelligence, Minneanapolis, MN, 1988.

....but most planning methods in use remain fairly insensitive to this information. Even in those efforts, the focus is on optimizing the non computational properties of the plan without regard to the computational resources required to do so. Work by Russell and Wefald [28, 29] and others [1, 7, 20, 21, 22, 23, 24] on rational allocation of computational effort in search has complemented this work on planning by studying optimal allocation of computation time in search, primarily in the simpler domain of game playing search. This work presumes a utility model for non search actions, and seeks to trade off ....

O. Hansson and A. Mayer. The optimality of satisficing solutions. In Proceedings of the Fourth Workshop on Uncertainty in Artificial Intelligence, 1988.

....These are often called probably correct algorithms. Some researchers, notably Valiant and others in the field of inductive learning [22, 47] have studied probably approximately correct algorithms, combining the above properties in the obvious way. However, as Horvitz [26] and Hansson and Mayer [19] have pointed out, what is needed is a theory of algorithms that maximize the comprehensive value of computation. In other words, since the utility of a computation and resulting action is a function of both the quality of the resulting solution and the time taken to choose it, we would like ....

....our own on the control of search. Heckerman and Jimison [23] have also used medical decision making as an example domain, showing how to vary the depth of analysis of a therapy problem according to its expected benefits. A third independent project was started in 1986 by Hansson and Mayer [18, 19, 20, 21], who proposed the use of information value as a means of controlling heuristic search, which in turn is implemented as probabilistic inference using information from the heuristic function as evidence. Dean s work on real time planning [7, 8] assumes a known variation of the intrinsic utility of ....

Hansson, O., and Mayer, A. (1988) The Optimality of Satisficing Solutions. Proceedings of the Fourth Workshop on Uncertainty in Artificial Intelligence, Minneapolis, MN, 1988.

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