Horvitz, E. J., and A. C. Klein (1993). "Utilitybased abstraction and categorization". In The Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence, Washington, DC, pp. 128-135  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Jensen et al. [10] and Shafer and Shenoy [20] Unfortunately, there are applications that CTP cannot deal with or where it is too slow (e.g. 18] Much recent effort has been spent on speeding up inference. The efforts can be classified into those that approximate (e.g. 15] 2] 9] 6] [7], 17] 22] 11] and [19] and those that exploit structures in the probability tables (e.g. 3] 1] We are interested in exploiting structures in the probability tables induced by independence of causal influence (ICI) The concept of ICI was first introduced by Heckerman [3] under the ....

E. J. Horvitz, and A. C. Klein (1993), Utility-based abstraction and categorization. in Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence, pp. 128-135.

....inconsistent information states and insignificant details before constructing a policy for the problem. The significance of the details in the information state is measured in terms of the effects of the information state on the posterior probabilities of (unobservable) state variables. Horvitz and Klein [1993] describe a decision theoretic approach to categorization based on utility. By aggregating states with similar utility values, and actions with similar values, decision models can be simplified for increased efficiency. Poh and Horvitz [1993] presents a greedy approach to exploring how random ....

....of (unobservable) state variables. Horvitz and Klein [1993] describe a decision theoretic approach to categorization based on utility. By aggregating states with similar utility values, and actions with similar values, decision models can be simplified for increased efficiency. Poh and Horvitz [1993] presents a greedy approach to exploring how random variables in a decision model might be refined, i.e. how they can be given a more fine grained set of values, to increase the utility of a decision. This work is intended to automate some of the effort that a decision analyst would put into ....

Horvitz, E. J., and Klein, A. C. 1993. Utility--based abstraction and categorization. In Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence, 128--135.

....The costs of building the decision tree are not taken into account; the costs of using the decision tree is compared to the cost of other on line approaches. In contrast, the information refinement approach is primarily concerned with the cost of constructing a decision tree. Horvitz and Klein [19] describe a decision theoretic approach to categorization based on the of utility of states or actions comprising the category. By aggregating states with similar utility values, and actions with similar values, decision 112 models can be simplified for increased efficiency. This approach is the ....

....details before constructing a policy for the problem. The significance of the details in the information state is measured in terms of the effects of the information state on the posterior probabilities of (unobservable) state variables. This is similar to the approach taken by Horvitz and Klein [19], who measure significance by the effects of information state on the decision maker s expected utility. Both of these approaches construct their abstractions during a preprocessing stage to decision making. This (and similar) preprocessing is still available for use before policies are ....

Eric J. Horvitz and Adrian C. Klein. Utility--based abstraction and categorization. In Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence, pages 128--135, 1993.

....inconsistent information states and insignificant details before constructing a policy for the problem. The significance of the details in the information state is measured in terms of the effects of the information state on the posterior probabilities of (unobservable) state variables. Horvitz and Klein [1993] describe a decision theoretic approach to categorization based on utility. By aggregating states with similar utility values, and actions with similar values, decision models can be simplified for increased efficiency. Poh and Horvitz [1993] presents a greedy approach to exploring how random ....

....of (unobservable) state variables. Horvitz and Klein [1993] describe a decision theoretic approach to categorization based on utility. By aggregating states with similar utility values, and actions with similar values, decision models can be simplified for increased efficiency. Poh and Horvitz [1993] presents a greedy approach to exploring how random variables in a decision model might be refined, i.e. how they can be given a more fine grained set of values, to increase the utility of a decision. This work is intended to automate some of the effort that a decision analyst would put into ....

Horvitz, E. J., and Klein, A. C. 1993. Utility--based abstraction and categorization. In Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence, 128--135.

.... s 2 e s 2 e S , fi fi fi e V e (es) Gamma V (s) fi fi fi ffi 2(1 Gamma fi) Proof We prove inductively that for all n fi fi fi e V n e (es) Gamma V n (s) fi fi fi n X i=0 ffi 2 fi i 14 The use of utility spans to generate abstractions is proposed by Horvitz and Klein [24], who use the notion in single step decision making. Our analysis can be applied to their framework to establish bounds on the degree to which an abstract decision can be less than optimal. Furthermore, the notion is useful in more general circumstances, as our results illustrate. Since V ....

Eric J. Horvitz and Adrian C. Klein. Utility-based abstraction and categorization. In Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence, pages 128--135, Washington, D.C., 1993.

....of this maximal span is exponential in the number of immediately relevant atoms, IR will always be restricted to atoms mentioned in the reward function R, which will be a rather small subset of atoms. The idea of using utility spans to generate abstractions is proposed by Horvitz and Klein [6], who use the notion in single step decision making. Our analysis can be applied to their framework to establish bounds on the degree to which an abstract decision can be less than optimal. Furthermore, the notion is useful in more general circumstances, as our results illustrate. 5 Concluding ....

Eric J. Horvitz and Adrian C. Klein. Utility-based abstraction and categorization. In Proc. of UAI-93, pages 128--135, Washington, D.C., 1993.

....we intend to carry out further analysis to seek a theoretic foundation for the current observation, and see how we may simplify the REMB function while maintaining satisfactory performance file of the algorithm. In addition, we will explore the utilization of the concept of value of information [7, 15] which is what really counts in deliberation scheduling. That is, the purpose of conducting more computation is to increase the value of information associated with the results. Thus, the goodness of a superstate should include its potential of increasing the value of information, not just that of ....

Horvitz, E. J., and A. C. Klein. Utility-based abstraction and categorization. Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence, Washington, DC, Morgan Kaufmann, pp. 128-135, 1993.

....slice. Knowing this will allow us to determine useful abstract views. Constructing abstract views based on this kind of sensitivity analysis does not, unfortunately, guarantee that the model of highest utility of a given size will be generated. Horvitz and Klein s work on utility based abstraction [Horvitz and Klein, 1993] provides a much more powerful method of generating abstractions. However, they assume that they can solve the entire problem with different abstractions, and then choose the one that has the highest utility. We are working in such large domains that the abstraction must be done before the problem ....

Horvitz, Eric J. and Klein, Adrian C. 1993. Utility-based abstraction and categorization. In Proceedings of the Ninth International Conference on Uncertainty in Artificial Intelligence, Washington, DC.

....by replacing a sequence of actions with an aggregate action the effect of which is consistent with the end effect of the action sequence. This amounts to doing sequential abstraction in our framework. Their work, however, did not offer any formal abstraction theory or concrete application. Horvitz [ 10 ] also discussed utility based state and action abstraction. In his framework abstraction is approximate, and utility is action based, instead of planbased. Boutilier et al. 1 ] proposed an approximate abstraction method for Markov Decision Processes. In his method the complexity of the domain ....

E. Horvitz. Utility-based abstraction and categorization. In Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence, pages 128--135, 1993.

....value is used determine the weights on these new attributes which are added to the relevant set. The process is continued until all the action descriptions have been examined. These weights can then be used by a clustering algorithm to generate good groupings of action branches. Horvitz [9] also discussed utility based state and action abstraction. In his framework, however, utility is action based, instead of plan based. Projection When abstraction is discussed in an uncertainty representation more expressive than the single probability distribution scheme, such as ours, many ....

E. Horvitz. Utility-based abstraction and categorization. In Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence, pages 128--135, 1993.

....and factors affecting the decisions, and assessing the utility of all the outcomes of our decisions. Utility theory has been used in such diverse domains as graphics rendering [22] display of information for time critical decision making [21] prioritization of repairs [8] and categorization [23]. 3.1 Approach Brown, et al. state that to ascribe user intent, we must identify the salient characteristics of our domain environment and specifically determine goals a user is trying to achieve and the actions to achieve those goals [10] Social scientists use intentions (as determined by ....

Eric J. Horvitz and Adrian C. Klein. Utility-based abstraction and categorization. In David Heckerman and Abe Mamdani, editors, Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence. Morgan Kaufmann, 1993.

.... networks have also been proposed, involving deletion of nodes, links, or states (Jensen and Andersen 1990; Kj rulff 1993; Sarkar 1993) The uncertain reasoning literature has also seen some alternate approaches to abstraction in probabilistic networks (Poh et al. 1994) and probabilistic reasoning (Horvitz and Klein 1993). Finally, there has been some investigation of the general problem of reasoning about the quality of approximate models (Laskey and Lehner 1994) We are currently exploring the relations among all these approaches and our own work. 6. CONCLUSION The foregoing experience suggests that iterative ....

Horvitz, E. J., and A. C. Klein (1993). Utility-based abstraction and categorization. Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence , Washington, DC, Morgan Kaufmann.

No context found.

Horvitz, E. J., and A. C. Klein (1993). "Utilitybased abstraction and categorization". In The Proceedings of the Ninth Conference on Uncertainty in Artificial Intelligence, Washington, DC, pp. 128-135

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