Kettler, B. P., Hendler, J. A., Andersen, W. A., and Evett, M. P. 1994. Massively Parallel Support for Case-Based Planning. IEEE Expert, February 1994, 8-14.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....1994] is another successful implementation of derivational replay in SNLP. Several other systems, provide specificstrong approaches to aspects of the case based planning paradigm. For example, Deja vu [Smyth and Keane, 1995] shows how retrieval can use a prediction of the adaptation cost, PARKA [Kettler et al. 1994] demonstrates massively parallel effective invocations to case memory during planning, and [L opez and Plaza, 1993] views medical diagnosis as a reactive planning task. One of the interesting and less explored (or explained) aspects of the case based planning paradigm is the use of multiple plans ....

B. P. Kettler, J. A. Hendler, A. W. Andersen, and M. P. Evett. Massively parallel support for case-based planning. IEEE Expert, 2:8--14, 1994.

.... in instance based learning (Aha et al. 1991) memory based (Stanfill and Waltz, 1986) or examplat based (Porter et al. 1990) Only some of the work has been applied to planning, usually in the context of analogy or case based reasoning (Hammond, 1989, Hanks and Weld, 1995, Kambhampati, 1989, Kettler et al. 1994, Veloso, 1994a, Veloso, 1994b) Most of this work concern domain specific algorithms (Hammond, 1989) Also, although these approaches demonstrated some useful lazy learning behavior, they did not, as we have, compare lazy and eager learning modes. The two main differences with these approaches ....

B. P. Kettler, J. A. Hendler, A. W. Andersen, and M.P. Evett. Massively parallel support for case-based planning. IEEE Expert, 2:8-14, 1994.

.... issue of testing the new plan for success and repairing it if a 8 CBP References Application Domain ASP II [ Alexander and Tsatsoulis, 1991 ] Diagnosis Therapy BOLERO [ L opez and Plaza, 1993; L opez and Plaza, 1995 ] Diagnosis Therapy CAESAR [ Fouque and Matwin, 1993] Software Reuse CAPER [ Kettler et al. 1994; Sanders et al. 1995 ] Logistics CAPLAN CBC [ Mu noz Avila and Huellen, 1995; Mu noz Avila and Huellen, 1996; Veloso et al. 1996] Process Planning CARS [ Fong and Yang, 1996 ] Software Reuse CHARADE [ Perini and Ricci, 1996; Ricci and Avesani, 1995 ] Diagnosis Therapy CHEF [ Hammond, ....

....(e.g. Branting, 1992; Jurisica, 1993 ] In a relational representation any feature of a case can be in relation with different features or different cases. The most common relations are abstraction specialization relations (e.g. is a) and partonomic relations (e.g. part of) For example CAPER [ Kettler et al. 1994 ] organizes both the episodic and the general knowledge in a semantic network. Figure 5 depicts an example of case in CAPER (the case case build car 1) where the case features (place, time, goal, and plan) are complex concepts 12 CBP Plan Representation Memory Organization ASP II ....

[Article contains additional citation context not shown here]

B. P. Kettler, J. A. Hendler, W. A. Andersen, and M. P. Evett. Massively parallel support for case-based planning. IEEE Expert, pages 8--14, February 1994.

.... systems, resorting to a knowledge base (usually of a large size) containing previously stored problems and solution failure plans in order to recall similar problem situations, is a frequent operation that, if implemented inefficiently, has a dramatic impact on the performance of the system [15]. WINs support the design of NC algorithms for the above problems. The class NC is the class of problems that are amenable to efficient parallel solutions, that is solutions that use a polynomial number of processors and take time polynomial in the logarithm of the input size (see [14] Problems ....

B. Kettler, J. Hendler, W. Andersen, and M. Evett, Massively Parallel Support for Case-based Planning, Proceedings of the 9th CAIA, Florida, 3--8, 1993.

....type retrieval are as follows: Greene (1991) proposes a neural network which replaces a hash function. The outputs of the neural network are a binary representation of a selected memory address. Such a scheme withstands noise and missing features since similar inputs produce similar results. Kettler et al. (1994) have constructed a case based system (CAPER) for matching structured plans. The system which is written in the PARKA language (Kettler et al., 1995) runs on the connection machine and plans are stored as semantic networks where each node relates to a concept and each connection to a ....

KETTLER B.P., HENDLER J.A., ANDERSEN W.A. and EVETT M.P. (1994) Massively Parallel Support for Case-Based Planning, IEEE Expert, Vol 9, No 1, pp 8-14.

.... in instance based (Aha et al. 1991) memory based (Stanfill and Waltz, 1986) or exemplarbased learning (Porter et al. 1990) Only some of the work has been applied to planning, usually in the context of analogy or case based reasoning (Hammond, 1989; Hanks and Weld, 1995; Kambhampati, 1989; Kettler et al. 1994; Veloso, 1994b; Veloso, 1994a) Most of this work concerns domain specific algorithms. Also, although these approaches demonstrated some useful lazy learning behavior, they did not, as we have, compare lazy and eager learning modes. The two main differences with these approaches are: control ....

Kettler, B. P., Hendler, J. A., Andersen, A. W., and Evett, M. P. (1994). Massively parallel support for case-based planning. IEEE Expert, 2:8--14.

....Wilkins 1988) Recently due to the developments being made in the areas of cognitive science and machine learning, designing planning systems that use case based approaches has become an attractive option. Many systems have been built (Kolodner 1987, Hammond 1989, Redmond 1990, Subbarao 1992, Kettler et al. 1994) within the framework of memory organization packets(MOPS) Schank 1982, Riesbeck 1989) In all these systems, entire plans are stored as cases to be retrieved in toto. A new plan is created by adapting a suitable plan or pieces of some plans merged together from memory to achieve the given goal. ....

....achieve the given goal. The issues of storing newly created plans in memory and the retrieval of plans from memory are the key factors for these systems. Learning is an other important characteristic for a CBP system. Present CBP systems (Kolodner 1987, Hammond 1989, Redmond 1990, Subbarao 1992, Kettler et al. 1994) have treated each of the following issues independently and less emphasis has been given in developing a unified system. i) Cost of plan modification (Hammond 1989) Adapting a plan or merging the pieces that needs less modification cost. ii) Hierarchical planning (Sacerdoti 1974, Tate 1975, ....

Kettler, B.P. et al. 1994. Massively Parallel Support for Case-Based Planning. IEEE EXPERT, February.

....which has not yet been extensively investigated even though many large databases employ relational representations. Data mining requires attention to efficiency concerns, which could spur research on designing typicality guided retrieval strategies (Porter et al. 1990) parallel case retrieval (Kettler et al. 1994), and software support for case construction (Kitano et al. 1993) Also needed are techniques for validating the behavior of case revision operators, and further research on how abstraction can be used to reduce search without requiring that solutions be downward refinable (Branting Aha, 1995) ....

Kettler, B. P., Hendler, J. A., Andersen, W. A., & Evett, M. P. (1994). Massively parallel support for case-based planning. IEEE Expert, 2, 8--14.

....abstraction in developing a CBR application, while the implementation of the language manages the indexing and retrieval. A way to improve efficiency is using parallelism, as done in the Parka language implemented in the Connection Machine, that uses a parallel algorithm for structure matching [Kettler] This approach is compatible with our formalization, but in our experience a good indexing scheme offers enough efficiency in retrieval for several realistic domains. The main goal of this paper has been to show that feature terms are a useful formalization that can be used to describe parts of ....

B P Kettler, J A Hendler, W A Anderson, M P Evett (1994), Massively parallel support for case-based planning. IEEE Expert, p. 8-14, Fed. 1994.

....1994] is another successful implementation of derivational replay in SNLP. Several other systems, provide specific strong approaches to aspects of the case based planning paradigm. For example, Deja vu [Smyth and Keane, 1995] shows how retrieval can use a prediction of the adaptation cost, PARKA [Kettler et al. 1994] demonstrates massively parallel effective invocations to case memory during planning, and [L opez and Plaza, 1993] views medical diagnosis as a reactive planning task. One of the interesting and less explored (or explained) aspects of the case based planning paradigm is the use of multiple plans ....

B. P. Kettler, J. A. Hendler, A. W. Andersen, and M. P. Evett. Massively parallel support for case-based planning. IEEE Expert, 2:8--14, 1994.

.... Although many CBR systems base retrieval on carefully constructed indexing vocabularies and problem descriptions, in order for retrieval to zero in on a small subset of the case library, other approaches exploit parallel hardware to maintain quick retrieval while considering large sets of cases (Kettler et al. 1994; Kolodner 1988b; Owens 1991; Stanfill Waltz 1986) Retrieval based on nontraditional types of input information, such as bitmap images and CAD plans, is also being investigated (Vo 1994) As is discussed in Section 5.4, methods are also being developed make retrieval focus on cases that are ....

Kettler, B.; Hendler, J.; Andersen, W.; and Evett, M. 1994. Massively parallel support for case-based planning. IEEE Expert 9(1):8--14.

.... in instance based learning (Aha et al. 1991) memory based (Stanfill and Waltz, 1986) or examplar based (Porter et al. 1990) Only some of the work has been applied to planning, usually in the context of analogy or case based reasoning (Hammond, 1989, Hanks and Weld, 1995, Kambhampati, 1989, Kettler et al. 1994, Veloso, 1994a, Veloso, 1994b) Most of this work concern domain specific algorithms (Hammond, 1989) Also, although these approaches demonstrated some useful lazy learning behavior, they did not, as we have, compare lazy and eager learning modes. The two main differences with these approaches ....

B. P. Kettler, J. A. Hendler, A. W. Andersen, and M. P. Evett. Massively parallel support for case-based planning. IEEE Expert, 2:8--14, 1994.

....can be used; and memory can be probed in numerous alternate ways, allowing more specific retrieval of stored plans that better fit a target problem with less adaptation. Preliminary version of an article appearing in IEEE Expert, February 1994, pp. 8 14. This paper is an extended version of [1]. 1 Introduction The CaPER system is a case based planner which is being developed to take advantage of the efficiency of plan reuse. It also addresses some of the problems and limitations of case based planners that use serial retrieval procedures on an indexed memory (casebase) CaPER makes ....

....The parallel procedures, in contrast, exhibited times growing better than logarithmically in the size of the casebase. Thus, using parallel methods, casebases in the thousands of cases can be accessed efficiently. Results similar to those above were also found for the automobile assembly domain [1]. While a serial implementation using an pre indexed memory might perform better than our serial implementation O(log n) in the number of cases if a balanced discrimination tree were used, such a system would suffer from the problems described in Section 1.2. 5 Related Work Closely related to ....

B.P. Kettler, J.A. Hendler, W.A. Andersen, and M.P. Evett, "Massively Parallel Support for Case-Based Planning," Proceedings of the Ninth Conference on Artificial Intelligence Applications (IEEE). Los Alamitos, California: IEEE Computer Society Press: 1993, pp. 3-11.

....a variety of queries that resulted in the retrieval of one or more plans (or parts of plans) that were combined to solve the target planning problem. These plans could be merged and harmful interactions among them resolved using annotations on a plan to capture interdependencies among its actions [8, 7, 5, 9]. Transportation Logistics Planning The United States Transportation Command (USTRANSCOM) is responsible for generating and maintaining the plans by which United States military forces are deployed. This responsibility includes determining the transportation needs for missions short and long, ....

Kettler, B.P., Hendler, J.A., Andersen, W.A., Evett, M.P., "Massively Parallel Support for Case-based Planning", IEEE Expert, Feb, 1994.

....for efficient matching. The structure matcher imposes no such restriction on the form of representation allowed. Cases to be efficiently retrieved from a large casebase that is not pre indexed while maintaining a rich relational representation. This algorithm is currently being evaluated in CaPER [12], a case based planning system designed to exploit massively parallel case retrieval techniques. 2 Motivation Most CBR systems pre index case memory. Pre indexing is the designation, at case storage time, of features ( indices ) through which a case may subsequently be retrieved. In serial ....

....to 1600 milliseconds on the larger casebase. As a point of comparison, an optimized serial version of this algorithm was implemented. The parallel algorithm running on the CM 2 achieved speedups ranging from 100 to 10; 000 over our serial version on representative CaPER queries 7 . Also, in [12], we show that the serial implementation of the structure matcher scales supralinearly in the size of the KB. We expect the parallel structure matcher s performance to scale (for reasons given in Section 6) to larger knowledge bases provided the size of the machine is not less than the size of the ....

[Article contains additional citation context not shown here]

Kettler, B.P., Hendler, J.A., Andersen, W.A., Evett, M.P., "Massively Parallel Support for Case-based Planning", IEEE Expert, 1993 (forthcoming).

....(Kambhampati Hendler, 1992) and extended to include information about failed planning decisions (Veloso, 1994) Recording the failed decisions may yield efficiency gains during adaptation. Table 1 compares the case based planners CHEF (Hammond, 1986) Priar (Kambhampati Hendler, 1992) CaPER (Kettler et al., 1994), Prodigy Analogy (Veloso, 1994) SPA (Hanks Weld, 1995) MPA (Francis Ram, 1995) derSNLP EBL (Ihrig Kambhampati, 1996) MRL (Koehler, 1994) Paris (Bergmann Wilke, 1995) and CAPlan CbC (Mu noz, 1998) according to the following criteria: ffl Adaptation Method: Many plan adaptation ....

.... a case is always the same, or dynamic, where similarity may change after problem solving episodes have completed (i.e. because learning mechanisms adjusted retrieval parameters) This similarity is usually precomputed and stored using some kind of indexing scheme, although the CaPER CBP system (Kettler et al. 1994) used an efficient retrieval scheme to compute similarity at run time based on the needs of the particular problem solving episode. ffl Search Space: Two primary planning paradigms state space and plan space, have been developed to search for solution plans. They mainly differ in how they ....

[Article contains additional citation context not shown here]

Kettler, B. P., Hendler, J. A., Andersen, W. A., & Evett, M. P. (1994). Massively parallel support for case-based planning. IEEE Expert, 2, 8--14.

....and previous successful cases of a given strategy) and if so, also uses the lattice to generate hypo style arguments for and against the success of a plan. caper is a case based planning system that makes use of massive parallelism to access a large casebase (currently several hundred cases) [17, 15]. caper uses structure based representations for cases and conceptual knowledge, implemented as a single semantic network. Given the availability of fast parallel case retrieval methods, memory does not have to be pre indexed for efficiency and thus can be accessed flexibly and often. 1 ....

....chaser indexes tort cases under features such as the harm caused and possible legal defenses [6] The main disadvantage of indexing is that it hinders flexibility at case retrieval time. Cases that share unindexed features with the target problem will not be retrieved. See discussion in [22, 17, 15]) A few systems have used parallel techniques to reduce the cost of matching. PARADYME, for example, is a massively parallel frame system that has been used to implement a memory for a CBR system [19] 2 In caper, the massively parallel mechanisms of the Parka Structure Matcher are used to ....

[Article contains additional citation context not shown here]

B. P. Kettler, J. A. Hendler, W. A. Andersen, and M. P. Evett. Massively parallel support for case-based planning. IEEE Expert, pages 8--14, Feb. 1994.

....to such queries as top down , and they are the bane of most serial KR systems, requiring O(n) time to effect, where n is the size of the KB. Serial systems use indexing schemes to mitigate this computational morass, but indexing can be unsatisfactory for a variety of reasons (as we discuss in [11]) including that typically it is infeasible to explicitly index all properties. The comparison between serial and parallel run times is more striking when realizing that for realistic networks d log(n) It is commonly believed that such network shallowness will persist and probably be ....

....by analysis. This performance compares very favorably with recognition queries on serial systems, which require O(pn) time for the same queries, where n is the size of the KB. Recognition queries in PARKA are independent of KB size, and should scale up to arbitrarily larger domains. Indeed, in [11] we report sub second run time performance of recognition queries in a case based planning system using KB s of over 10,000 frames. We achieved run time performance 10,000 times better than that of a highly optimized serial version of PARKA 6 . This experiment was designed not only to ....

[Article contains additional citation context not shown here]

B. Kettler, J.A. Hendler, W.A. Andersen, and M.P. Evett. Massively Parallel Support for Case-based Planning, IEEE Expert, 1993 (forthcoming).

....and previous successful cases of a given strategy) and if so, also uses the lattice to generate hypo style arguments for and against the success of a plan. caper is a case based planning system that makes use of massive parallelism to access a large casebase (currently several hundred cases) [16, 15]. caper uses graph structured representations for cases and conceptual knowledge, implemented as a single semantic network. Given the availability of fast parallel case retrieval methods, memory does not have to be pre indexed for efficiency and thus can be accessed flexibly and often. 4 ....

....chaser indexes tort cases under features such as the harm caused and possible legal defenses [7] The main disadvantage of indexing is that it hinders flexibility at case retrieval time. Cases that share unindexed features with the target problem will not be retrieved. See discussion in [20, 16, 15]) A few systems have used parallel techniques to reduce the cost of matching. paradyme, for example, is a massively parallel frame system that has been used to implement a memory for a CBR system [18] 6 In caper, the massively parallel mechanisms of the Parka Structure Matcher are used to ....

B. P. Kettler, J. A. Hendler, W. A. Andersen, and M. P. Evett. Massively parallel support for case-based planning. IEEE Expert, pages 8--14, Feb. 1994.

....The detailed set of operators provides long plans ( 40 steps) with many possible solutions to the same problem, and thus this domain can also be used to evaluate the solution quality of planning systems. UM Translog is currently being used in the evaluation of a case based planning system, CaPER [6, 5], and a hierarchical task network planning system, UMCP [3, 2] It is also being used by UM Nonlin [4] a common lisp implementation of Austin Tate s Nonlin HTN planning system, to generate a plan library for plan reuse by CaPER. Due to the complexity and size of the domain, it is not easy to ....

Brian P. Kettler, James A. Hendler, William A. Andersen, and Matthew P. Evett. Massively parallel support for case-based planning. IEEE Expert, pages 8--14, February 1994.

....the serial client simulator and the parallel server. 2 The original Parka language and its use for knowledge representation is described in [5, 4] These documents are largely superseded by this document. Several Parka applications are being built including a case based planning system, CaPER [3] and a hybrid knowledge base database system. Mailing List A Parka mailing list exists (parka users cs.umd.edu) To be added to this list, send email to parka users request cs.umd.edu. Worldwide Web A Parka Worldwide Web (WWW) page has been set up at URL ....

Brian P. Kettler, James A. Hendler, William A. Andersen, and Matthew P. Evett. Massively parallel support for case-based planning. IEEE Expert, pages 8--14, February `1994.

No context found.

Kettler, B. P., Hendler, J. A., Andersen, W. A., and Evett, M. P. 1994. Massively Parallel Support for Case-Based Planning. IEEE Expert, February 1994, 8-14.

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