Manuela M. Veloso and Jaime G. Carbonell. Towards scaling up machine learning: A case study with derivational analogy in prodigy. In Maria Zemankova and Zbigniew W. Ras, editors, Machine Learning Methods for Planning, pages 233--272. Morgan Kaufmann, San Mateo, ca, 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....is also based on proof generalization introducing meta variables in proofs to nd a minimal proof using given rule schemata. Melis [13] applied the paradigm of derivational analogy to theorem proving at the more abstract planning level. The notion of derivational analogy was invented by Carbonell [4, 19] for planning and general problem solving where the lines of reasoning (i.e. the sequence of decisions and their justi cations) are replayed to solve a target problem. Melis approach is based on the paradigm of proof planning and thus on an explicit representation of control knowledge in terms ....

M. Veloso and J.G. Carbonell. Towards scaling up machine learning: A case study with derivational analogy in prodigy. In S. Minton, editor, Machine Learning Methods for Planning. M. Kaufmann, 1993.

....we named s into tw (for current tower ) for more intuitive readability. Remember, that the situation variable s is introduced to represent the current situation (state) which is returned if the predicates at the parent node are true in this state. 4.1. 2 Unloading and loading objects Veloso [VC93a] proposed the rocket and a more general transportation domain. Transportation problems typically involve the loading of objects into a transport vehicle and their unloading at a given destination. Plan construction for transportation problems often relies on interleaving of subgoals to generate ....

....Binary tree for unload Figure 4.7: Binary tree for load 52 CHAPTER 4. TRANSFORMATION OF PLANS TO PROGRAMS 4.2 Complex cyclic macros: non linear structures Examples for non linear structures are tower (presented in chapter 3. 2 and discussed in [Wys87] rocket and other transportation domains [VC93a] and hanoi. A non linear D plan not necessarily implies that the resulting cyclic macro has to be non linear. The reason for this can be found in function theory [FH88] A given syntactical realization of a function does not necessarily correspond with its semantic complexity [Hin78, Odi89] For ....

M. M. Veloso and J. G. Carbonell. Towards scaling up machine learning: A case study with derivational analogy in prodigy. In S. Minton, editor, Machine Learning Methods for Planning, chapter 8. Morgan Kaufmann, 1993. BIBLIOGRAPHY 85

....investigations in learning adaptation knowledge. Some approaches for learning adaptation knowledge can be found in DIAL (David Leake, 1993; Leake, 1995b; Leake, 1995a) or CHEF (Hammond, 1986; Hammond, 1989) These systems use knowledge intensive derivational analogy approaches (Carbonell, 1986; Veloso and Carbonell, 1993) to learn adaptation knowledge. Knowledge intensive means that these approaches require a lot of background and problem solving knowledge. For example in DIAL and CHEF 1 University of Kaiserslautern Centre for Learning Systems and Applications (LSA) Department of Computer Science P.O. Box 3049, ....

Veloso, M. M. and Carbonell, J. G. (1993). Toward scaling up machine learning: A case study with derivational analogy in prodigy. In Minton, S., editor, Machine Learning Methods for Planning, chapter 8, pages 233--272. Morgan Kaufmann, San Mateo.

.... Rosenbloom Laird, 1986; Minton, 1988; Minton, Carbonell, Knoblock, Kuokka, Etzioni, Gil, 1989; Shavlik O Rorke, 1993; Etzioni, 1993; Minton Zweben, 1993; Langley Allen, 1993; Kambhampati Kedar, 1994) and analogical or case based reasoning (Carbonell, 1986; Kambhampati Hendler, 1992; Veloso Carbonell, 1993; Veloso, 1994) As the main contribution of this paper, we present an abstraction methodology and a related learning method in which beneficial abstract planning cases are automatically derived from given concrete cases. Based on a given concrete and abstract language, this learning approach ....

....with a related solution. As is the case in Prodigy (Minton et al. 1989) we only consider sequential plans, i.e. plans with totally ordered operators. The planning cases we assume do not include a problem solving trace as for example the problem solving cases in Prodigy Analogy (Veloso, 1992; Veloso Carbonell, 1993; Veloso, 1994) In real world applications, a domain expert s solutions to previous problems are usually recorded in a company s filing cabinet or database. These cases can be seen as a collection of the company s experience, from which we want to draw power. During a learning phase, a set of ....

[Article contains additional citation context not shown here]

Veloso, M. M., & Carbonell, J. G. (1993). Towards scaling up machine learning: A case study with derivational analogy in PRODIGY. In Minton, S. (Ed.), Machine Learning Methods for Planning, chap. 8, pp. 233--272. Morgan Kaufmann.

....investigations in learning adaptation knowledge. Some approaches for learning adaptation knowledge can be found in DIAL (David Leake, 1993; Leake, 1995b; Leake, 1995a) or CHEF (Hammond, 1986; Hammond, 1989) These systems use knowledge intensive derivational analogy approaches (Carbonell, 1986; Veloso and Carbonell, 1993) to learn adaptation knowledge. Knowledge intensive means that these approaches require a lot of background and problem solving knowledge. For example in DIAL and CHEF adaptation strategies for special problem fields are acquired based on general domain knowledge. So a reduction of the knowledge ....

Veloso, M. M. and Carbonell, J. G. (1993). Toward scaling up machine learning: A case study with derivational analogy in prodigy. In Minton, S., editor, Machine Learning Methods for Planning, chapter 8, pages 233--272. Morgan Kaufmann, San Mateo.

....to previous work in AI. In particular, there is a long history of AI programs that use repair or debugging strategies to solve problems, primarily in the areas of planning and design[38, 41] This approach has recently had a renaissance with the emergence of casebased [11, 24] and analogical [14, 22, 43] problem solving. To solve a problem, a case based system will retreive the solution from a previous, similar problem and repair the old solution so that it solves the new problem. There has also been related work in AI on sophisticated methods for measuring the contention between resources in ....

M.M. Veloso and J.G. Carbonell. Towards scaling up machine learning: A case study with derivation analogy in prodigy. In Minton S., editor, Machine Learning Methods for Planning and Scheduling. Morgan Kaufmann, 1992.

....is derived (this would require problem solving knowledge) but that a solutions solves the given problem, i.e. a proof of the correctness of the solution. It is really important to keep this distinction in mind since it makes our approach different from derivational analogy [Carbonell, 1986, Veloso and Carbonell, 1993] On the other hand, the domain knowledge we employ for explaining a case is stronger than just causal relations like in other approaches [Barletta and Mark, 1988, Koton, 1988] Moreover, explanations based on strong domain knowledge can be easily derived automati cally and do not need to be ....

....Planning as Example Domain. To demonstrate the application of the explanation based similarity approach for case based planning, we present an example from the field of production planning in mechanical engineering adapted from the CabPlan System [Paulokat and Wess, 1993] a PRODIGYlike approach [Veloso and Carbonell, 1993]. The goal is to generate a process plan for the production of a rotationally symmetric workpiece on a lathe. The problem description, which may be derived from a CAD drawing, contains the complete specification (especially the geometry) of the desired workpiece (goal state) together with a ....

[Article contains additional citation context not shown here]

M. M. Veloso and J. G. Carbonell. Towards scaling up machine learning: A case study with derivational analogy in PRODIGY. In Steven Minton, editor, Machine Learning Methods for Planning, chapter 8, pages 233--272. Morgan Kaufmann, 1993.

....only parts of this object structure may be useful. Reusing only the top level objects of such a case may be comparable to reusing an abstract case as in Paris. 4. 4 Other approaches to case based planning Most similar to the PARIS approach are the case based planning systems Prodigy Analogy [10], PRIAR [6] and CaPlan CbC [9, 8] However, these systems reuse planning cases directly and without doing any abstraction. Our approach is also related to the idea of skeletal plans [5] In the skeletal plan approach no model of the operators (neither concrete nor abstract) is used to describe ....

M. M. Veloso and J. G. Carbonell, `Towards scaling up machine learning: A case study with derivational analogy in PRODIGY', in Machine Learning Methods for Planning, ed., Steven Minton, chapter 8, 233--272, Morgan Kaufmann, (1993). PARIS 5 R. Bergmann and W. Wilke

....general purpose planning. We demonstrate here that effect social laws can be learned for each new domain, using genetic algorithms or even more efficiently using hill climbing approaches. Additonal approaches could be considered to learning these laws. For example, analogical learning approaches [Coo91, VC93, Kam93] could be used to transfer application of a useful set of laws from one domain to other similar domains. Reinforcement learning approaches [Lin92, MC91] could also be used to assign credit to candidate laws in a new domain. 5 Conclusions Our research presented herein provides to the artificial ....

Manuela M. Veloso and Jaime G. Carbonell. Toward scaling up machine learning: a case study with derivational analogy in prodigy. In Steven Minton, editor, Machine Learning Methods for Planning, chapter 8, pages 233--272. Morgan Kaufmann, 1993.

....is required to compute state abstractions. Hence, the generic abstraction theory should not require complicated inferences and should avoid backtracking within the SLD refutation procedure. 8 Related Work Most similar to the PARIS approach are the case based planning systems Prodigy Analogy [ Veloso and Carbonell, 1993 ] and PRIAR [ Kambhampati and Hendler, 1992 ] However, these systems reuse planning cases directly and without doing any abstraction. Knoblock [ Knoblock, 1994 ] presented an approach to automatically constructing abstraction hierarchies. This approach is limited to abstraction by dropping ....

M. M. Veloso and J. G. Carbonell. Towards scaling up machine learning: A case study with derivational analogy in PRODIGY. In Steven Minton, editor, Machine Learning Methods for Planning, chapter 8, pages 233--272. Morgan Kaufmann, 1993.

....is derived (this would require problem solving knowledge) but that a solutions solves the given problem, i.e. a proof of the correctness of the solution. It is really important to keep this distinction in mind since it makes our approach different from derivational analogy [ Carbonell, 1986; Veloso and Carbonell, 1993 ] On the other hand, the domain knowledge we employ for explaining a case is stronger than just causal relations like in other approaches [ Barletta and Mark, 1988; Koton, 1988 ] Moreover, explanations based on strong domain knowledge can be easily derived automatically and do not need to ....

....as Example Domain. To demonstrate the application of the explanation based similarity approach for case based planning, we present an example from the field of production planning in mechanical engineering adapted from the CaPlan System [ Paulokat and Wess, 1993 ] a PRODIGYlike approach [ Veloso and Carbonell, 1993 ] The goal is to generate a process plan for the production of a rotationally symmetric workpiece on a lathe. The problem description, which may be derived from a CAD drawing, contains the complete specification (especially the geometry) of the desired workpiece (goal state) together with a ....

[Article contains additional citation context not shown here]

M. M. Veloso and J. G. Carbonell. Towards scaling up machine learning: A case study with derivational analogy in PRODIGY. In Steven Minton, editor, Machine Learning Methods for Planning, chapter 8, pages 233--272. Morgan Kaufmann, 1993.

....an abstract case very easily. This can open up the additional opportunity to involve the user in the planning process, for example in the selection of an abstract case she he favors. 7 Related Work Most similar to the PARIS approach are the case based planning systems Prodigy Analogy [ Veloso and Carbonell, 1993 ] and PRIAR [ Kambhampati and Hendler, 1992 ] However, these systems reuse planning cases directly and without doing any abstraction. Knoblock [ Knoblock, 1994 ] presented an approach to automatically constructing abstraction hierarchies. This approach is limited to abstraction by dropping ....

M. M. Veloso and J. G. Carbonell. Towards scaling up machine learning: A case study with derivational analogy in PRODIGY. In Steven Minton, editor, Machine Learning Methods for Planning, chapter 8, pages 233--272. Morgan Kaufmann, 1993.

....from a complex domain. Introduction Case based planning provides significant performance improvements over generative planning when the planner is solving a series of similar problems, and when it has an adequate theory of problem similarity (Hammond 1990; Ihrig 1996; Ihrig Kambhampati 1994; Veloso Carbonell 1993). One approach to case based planning is to store plan derivations which are then used as guidance when solving new similar problems (Veloso Carbonell 1993) Recently we adapted this approach, called derivational replay, to improve the performance of the partial order planner, snlp (Ihrig ....

.... a series of similar problems, and when it has an adequate theory of problem similarity (Hammond 1990; Ihrig 1996; Ihrig Kambhampati 1994; Veloso Carbonell 1993) One approach to case based planning is to store plan derivations which are then used as guidance when solving new similar problems (Veloso Carbonell 1993). Recently we adapted this approach, called derivational replay, to improve the performance of the partial order planner, snlp (Ihrig Kambhampati 1994) Although it was found that replay tends to improve overall performance, its effectiveness depends on retrieving an appropriate case. This ....

[Article contains additional citation context not shown here]

Veloso, M., and Carbonell, J. 1993. Toward scaling up machine learning: A case study with derivational analogy in prodigy.

....to previous work in AI. In particular, there is a long history of AI programs that use repair or debugging strategies to solve problems, primarily in the areas of planning and design[37, 40] This approach has recently had a renaissance with the emergence of case based[14, 26] and analogical [17, 24, 42] problem solving. To solve a problem, a case based system will retreive the solution from a previous, similar problem and repair the old solution so that it solves the new problem. The fact that the min conflicts approach performs well on n queens, a well studied, standard constraintsatisfaction ....

M.M. Veloso and J.G. Carbonell. Towards scaling up machine learning: A case study with derivation analogy in prodigy. In Minton S., editor, Machine Learning Methods for Planning and Scheduling. Morgan Kaufmann, 1992.

....according to the number of remaining uncovered goals and no more interactions of size no int goals are found. The retrieval effort may be interrupted by the procedure Stop Retrieval p when a threat is recognized in its potential benefits with respect to problem solving search savings (see [ Veloso and Carbonell, 1993b ] 5 Flexible Replay of Multiple Guiding Cases When a new problem is proposed, prodigy analogy retrieves from the case library one or more problem solving episodes that may partially cover the new problem solving situation. The system uses a similarity metric that weighs goalrelevant features ....

....] 5 Flexible Replay of Multiple Guiding Cases When a new problem is proposed, prodigy analogy retrieves from the case library one or more problem solving episodes that may partially cover the new problem solving situation. The system uses a similarity metric that weighs goalrelevant features [ Veloso and Carbonell, 1993b ] In a nutshell, it selects a set of past cases that solved subsets of the new goal statement. The initial state is partially matched in the features that were relevant to solving these goals in the past. Each retrieved case provides guidance to a set of interacting goals from the new goal ....

[Article contains additional citation context not shown here]

Manuela M. Veloso and Jaime G. Carbonell. Towards scaling up machine learning: A case study with derivational analogy in prodigy. In S. Minton, editor, Machine Learning Methods for Planning, pages 233--272. Morgan Kaufmann, 1993.

....derivational analogy algorithm used by the system. When a new problem is proposed, Prodigy Analogy retrieves from the case library one or more problem solving episodes that may partially cover the new problem solving situation. The system uses a similarity metric that weighs goal relevant features (Veloso Carbonell, 1993). Essentially, it selects a set of past cases that solved subsets of the new goal statement. The initial state is partially matched in the features that were relevant to solving these goals in the past. Each retrieved case provides guidance to a set of interacting goals from the new goal ....

Veloso, M., & Carbonell, J. G. (1993). Towards scaling up machine learning: A case study with derivational analogy in PRODIGY (pp. 233-272). In S. Minton (Ed.), Machine learning methods for planning. Morgan Kaufmann.

....derivational analogy algorithm used by the system. When a new problem is proposed, Prodigy Analogy retrieves from the case library one or more problem solving episodes that may partially cover the new problem solving situation. The system uses a similarity metric that weighs goal relevant features [14]. Essentially, it selects a set of past cases that solved subsets of the new goal statement. The initial state is partially matched in the features that were relevant to solving these goals in the past. Each retrieved case provides guidance to a set of interacting goals from the new goal ....

Veloso, M., & Carbonell, J. G. (1993). Towards scaling up machine learning: A case study with derivational analogy in PRODIGY (pp. 233-272). In S. Minton (Ed.), Machine learning methods for planning. Morgan Kaufmann.

....cases occurs by extending the base level planner with the ability to examine its internal decision cycle, recording the justifications for each decision during its search process. Prodigy Analogy has been re implemented in Prodigy4.0, a state space nonlinear planner (Carbonell et al. 1992; Fink Veloso 1994) Prodigy4.0 s planning reasoning cycle involves several decision points, namely: the goal to select from the set of pending goals; the operator to choose to achieve a particular goal; the bindings to choose in order to instantiate the chosen operator; apply an operator whose preconditions are satisfied or ....

Veloso, M. M., and Carbonell, J. G. 1993b. Towards scaling up machine learning: A case study with derivational analogy in prodigy. In Minton, S., ed., Machine Learning Methods for Planning. Morgan Kaufmann. 233--272.

No context found.

Veloso, M. M., and Carbonell, J. G. 1993b. Towards scaling up machine learning: A case study with derivational analogy in PRODIGY. In Minton, S., ed., Machine Learning Methods for Planning. Morgan Kaufmann. 233--272.

No context found.

Manuela M. Veloso and Jaime G. Carbonell. Towards scaling up machine learning: A case study with derivational analogy in prodigy. In Maria Zemankova and Zbigniew W. Ras, editors, Machine Learning Methods for Planning, pages 233--272. Morgan Kaufmann, San Mateo, ca, 1993.

No context found.

Veloso, M., & Carbonell, J. (1993b). Toward scaling up machine learning: A case study with derivational analogy in prodigy. In Minton, S. (Ed.), Machine Learning methods for planning. Morgan Kaufmann.

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