Robert B. Doorenbos and Manuela M. Veloso. Knowledge organiza- tion and the utility problem. In Proceedings of the Third International Workshop on Knowledge Compilation and Speedup Learning, pages 28-34, Amherst, MA, June 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....1988#. Research on the utility problem can be divided up into twokey issues. The #rst issue is the expensive rule problem in which individual learned rules are so expensive to match that the system su#ers a signi#cant slowdown from learning. The second issue is the averagegrowth e#ect #Doorenbos, 1993#, which results from the cumulative expense of having learned many rules. If the time required to eliminate from consideration all of the rules that are not relevant to a particular situation scales poorly with the total number of rules in the system, this could potentially lead to a signi#cant ....

....rules in the system, this could potentially lead to a signi#cant slowdown from learning. Fortunately, recent work on the average growth e#ect has shown that, by exploiting sharing and eliminating irrelevant match e#ort, it is possible to learn over one million rules with a sublinear cost increase #Doorenbos, 1993; Doorenbos, 1994#. 2 This leaves the expensive rule problem as the remaining open question, and thus what is focused on here. In this article we takeanovel approach to the expensive rule problem byinvestigating the idea that expensiveness is inadvertently and unnecessarily introduced into ....

[Article contains additional citation context not shown here]

Doorenbos, B. & Veloso, M. M. #1993#. Knowledge organization and the utility problem. In Proceedings of the Third International Workshop on Knowledge Compilation and Speedup Learning.

....to the bounded explained control rules. Nevertheless, we are currently developing efficient methods for organizing and matching the learned control rules. We consider this organization essential to the overall learning process to avoid a potential utility problem due to inefficient matching (Doorenbos and Veloso, 1993). In the case of not using the rules, the reason why it did not solve all problems was because the given time bound. But, after analyzing why that happened, we concluded that some rules were not correct after the training phase, that lead us to the next set of experiments. 5.3 Convergence to the ....

Robert B. Doorenbos and Manuela M. Veloso. Knowledge organiza- tion and the utility problem. In Proceedings of the Third International Workshop on Knowledge Compilation and Speedup Learning, pages 28-34, Amherst, MA, June 1993.

....rules that the time spent solving the problem degraded so much to consider it a utility problem [11] However, we are currently developing efficient methods for organizing and matching the learned control rules. We consider this organization essential and part of the overall learning process [4]. 5 Related work Most speedup learning systems have been applied to problem solvers with the linearity assumption, such as the ones applied to Prolog or logic programming problem solvers [15, 21] special purpose problem solvers [12, 9, 18] or other general purpose linear problem solvers [5, ....

Robert B. Doorenbos and Manuela M. Veloso. Knowledge organization and the utility problem. In Proceedings of the Third International Workshop on Knowl- edge Compilation and Speedup Learning, pages 28--34, Amherst, MA, June 1993.

....21.5 to 9.5 in the blocksworld, and from 19 to 8.3 in the logistics. This corresponds to a considerable increase in the solvability horizon of the problem solver when using the rules. Also, since the matcher for the control rules is not using any optimum retrieving and organization algorithm [7], the time spent matching the rules represents the usual utility problem. The results shown in the table are especially relevant as the use of the learned set of rules outperformed the base level problem solver even with the rudimentary matcher. 8 Related Work There are several dimensions along ....

Robert B. Doorenbos and Manuela M. Veloso. Knowledge organization and the utility problem. In Proceedings of the Third International Workshop on Knowledge Compilation and Speedup Learning, pages 28--34, Amherst, MA, June 1993.

....to the bounded explained control rules. Nevertheless, we are currently developing efficient methods for organizing and matching the learned control rules. We consider this organization essential to the overall learning process to avoid a potential utility problem due to inefficient matching (Doorenbos and Veloso, 1993). After analyzing why some problems were not solved, we concluded that some rules were not correct after the training phase. This fact led us to carry on the next set of experiments towards testing the convergence of the learning approach. 5.3 Convergence to the Correct Control Knowledge The ....

Doorenbos, R. B. and Veloso, M. M. (1993). Knowledge organization and the utility problem.

....of equal or shorter length in 92 of the problems. prodigy analogy includes an indexing mechanism for the case library of learned problem solving episodes [ Veloso and Carbonell, 1993b ] We verified that with this memory organization, we reduced (or avoided) the potential utility problem [ Doorenbos and Veloso, 1993 ] The retrieval time suffers no significant increase with the size of the case library. 0 100 200 300 400 500 600 700 800 900 1000 Number of Problems Solved 0 50 100 150 200 250 300 350 400 Time Bound (seconds) Prodigy Analogy NoLimit 0 50000 100000 150000 200000 250000 Cumulative Running ....

Robert B. Doorenbos and Manuela M. Veloso. Knowledge organization and the utility problem. In Proceedings of the Third International Workshop on Knowledge Compilation and Speedup Learning, pages 28--34, Amherst, MA, June 1993.

....to the bounded explained control rules. Nevertheless, we are currently developing efficient methods for organizing and matching the learned control rules. We consider this organization essential to the overall learning process to avoid a potential utility problem due to inefficient matching (Doorenbos and Veloso, 1993). In the case of not using the rules, the reason why it did not solve all problems was because the given time bound. But, after analyzing why that happened, we concluded that some rules were not correct after the training phase, that lead us to the next set of experiments. 5.3 Convergence to the ....

Robert B. Doorenbos and Manuela M. Veloso. Knowledge organization and the utility problem. In Proceedings of the Third International Workshop on Knowledge Compilation and Speedup Learning, pages 28--34, Amherst, MA, June 1993.

....solutions of equal or shorter length in 92 of the problems. PRODIGY ANALOGY includes an indexing mechanism for the case library of learned problem solving episodes (Veloso Carbonell 1993b) We verified that with this memory organization, we reduced (or avoided) the potential utility problem (Doorenbos Veloso 1993): The retrieval time suffers no significant increase with the size of the case library. Discussion and related work PRODIGY s problem solving method is a combination of means ends analysis, backward chaining, and state space search. PRODIGY commits to particular choices of operators, bindings, ....

Doorenbos, R. B., and Veloso, M. M. 1993. Knowledge organization and the utility problem. In Proceedings of the Third International Workshopon KnowledgeCompilation and Speedup Learning, 28--34.

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