Borrajo, D., and Veloso, M. 1994. Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94, 64--82. Springer Verlag.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....regressed back to the step that achieved the latter condition. The regressed condition became the precondition for when not to select that operator to work towards that goal . Unfortunately, the impossibility theory is domain specific and each new domain requires a new one. 2.2. 3 HAMLET HAMLET[4] learns PRODIGY select rules for the different types of choice points found in NOLIMIT[20] PRODIGY s non linear problem solver. HAMLET uses two mechanisms to produce overgeneral preconditions for these search control rules. One is to use an explanation template (that is specific for the type of ....

D. Borrajo and M. Veloso. Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, April 1994.

....EBL involves combining it with inductive learning methods. In particular, EBL methods can be used to isolate the features of the problem that are relevant to the failure and then inductive methods can be used to generalize over these partial explanations of failure (or success) Borrajo and Veloso [4] discuss an approach of this type in the context of a state space planner, while Estlin and Mooney present a similar method in the context of partial order planning [12] It would be interesting to see how such hybrid methods can be adapted to UCPOP EBL. 22 It is tempting to use the complete ....

D. Borrajo and M. Veloso. Incremental learning of control knowledge for nonlinear problem solving. In Proc. European Conference on Machine Learning, 1994.

....than EBL did alone. These results are particularly significant since UCPOP EBL utilizes additional domain axioms which were not provided to Scope. Most other related learning systems have been evaluated on different planning algorithms, thus system comparisons are difficult. The Hamlet system (Borrajo Veloso 1994) learns control knowledge for the nonlinear planner underlying Prodigy4.0. Hamlet acquires rules by explaining past planning decisions and then incrementally refining them. Since, Prodigy4.0 is not a partial order planner it is difficult to directly compare Hamlet and Scope. When making a rough ....

....planner underlying Prodigy4.0. Hamlet acquires rules by explaining past planning decisions and then incrementally refining them. Since, Prodigy4.0 is not a partial order planner it is difficult to directly compare Hamlet and Scope. When making a rough comparison to the results reported in Borrajo Veloso (1994b) Scope achieves a greater speedup factor in blocksworld (11.3 vs 1.8) and in the logistics domain (5.3 vs 1.8) Future Work There are several issues we hope to address in future research. First, replacing Foil s informationgain metric for picking literals with a metric that more directly ....

Borrajo, D., and Veloso, M. 1994. Incremental learning of control knowledge for nonlinear problem solving. In Proc.

.... Rosenblitt, 1991; Barrett Weld, 1993) and the efficiency of the planning process itself (e.g. Minton, 1988) Until now some initial work is done on acquiring and representing expertise and controlling a planner to support optimization of plan execution costs (e.g. P erez Carbonell, 1993; Borrajo Veloso, 1994)) A main characteristic of human planning for machining is the use of examples that have been found to be successful in similar situations (Humm, Schulz, Radtke, Warnecke, 1991) In mechanical engineering, there are numerous attempts to build up index structures to support the classification of ....

Borrajo, D., & Veloso, M. (1994). Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning.

....point. The problem solver has a very powerful language to express both domain and control knowledge. Learning Control Knowledge Control rules can be learned automatically by the system by static analysis of the domain operators (Etzioni 1990) analysis of problem solving traces (Minton 1988; Borrajo Veloso 1994), or a combination of both (Perez Etzioni 1992) In addition to learning control rules, PRODIGY can also control the search using derivational analogy with similar previously solved problems (Veloso Carbonell 1993) Search is also more efficient when PRODIGY is used as a hierarchical problem ....

....the process planning domain. These control rules are hand coded, and continue to grow in number as we continue to understand how to controlthe search complexity of the domain. Some of the work on automatically learning control knowledge in PRODIGY s has been applied to the process planning domain (Borrajo Veloso 1994). Learning to Generate Process Plans of Good Quality The performance of the planner can also be improved by learning new rules to guide the search towards better quality solutions. The mechanism to learn quality enhancing control knowledge described previously has been applied to the process ....

Borrajo, D., andVeloso, M. 1994. Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML94. Sicily, Italy: Springer Verlag.

....is a member of the target concept, and then confirms the conjecture with empirical data. Other systems have employed lazy explanation based learning (LEBL) which generates incomplete explanations and then incrementally refines any overly general knowledge using new examples (Tadepalli, 1989; Borrajo Veloso, 1994). This dissertation presents a novel multi strategy learning approach to control knowledge acquisition for planning systems. The Scope learning system also uses a combination of EBL and induction to learn control information. However, instead of generating control rules through EBL and then ....

Borrajo, D., & Veloso, M. (1994). Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94, pp.

....developing methods that automatically acquire search control knowledge from experience. However, most work in learning and planning has been in the context of linear, state based planners[14, 8, 12] More recently, the problem of learning search control for a nonlinear planner has been presented [3, 10, 18]. Nonlinear planners have been accepted as superior to linear planners for many years, and recent experimental results support that partial order planners are more efficient than totally ordered planners in most domains [1, 15, 9] Though some past work has addressed this problem [5, 7] there has ....

....results are particularly significant since SNLP EBL was used to learn control rules for several types of planning decisions while Dolphin was constrained to learn rules for only new action selection. SNLP EBL also requires additional domain knowledge which was not provided to Dolphin. Hamlet [3] is another related system which learns control knowledge for the nonlinear planner underlying Prodigy4.0. Hamlet al..so combines induction with EBL but uses a very different approach than Dolphin. It is diffcult to directly compare Hamlet and Dolphin for several reasons. First, Hamlet is directly ....

D. Borrajo and M. Veloso. Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94, pages 64--82, Springer Verlag, 1994.

No context found.

Borrajo, D., and Veloso, M. 1994. Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94, 64--82. Springer Verlag.

....positive example only if the decision leads to one of the globally best solutions and it was not the choice selected by the problem solving default heuristics. We implemented our learning approach in a system called HAMLET, standing for Heuristics Acquisition Method by Learning from search Trees (Borrajo and Veloso, 1994, Veloso and Borrajo, 1994) HAMLET is integrated with PRODIGY4.0, the current nonlinear problem solver of the PRODIGY architecture for planning and learning (Carbonell et al. 1992) HAM LET learns control knowledge incremental and inductively to improve both the search efficiency of the problem ....

....leads to one of the globally best solutions and it was not the choice selected by the problem solving default heuristics. We implemented our learning approach in a system called HAMLET, standing for Heuristics Acquisition Method by Learning from search Trees (Borrajo and Veloso, 1994, Veloso and Borrajo, 1994). HAMLET is integrated with PRODIGY4.0, the current nonlinear problem solver of the PRODIGY architecture for planning and learning (Carbonell et al. 1992) HAM LET learns control knowledge incremental and inductively to improve both the search efficiency of the problem solver and to improve the ....

[Article contains additional citation context not shown here]

Daniel Borrajo and Manuela M. Veloso. Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94, pages 64-82. Springer Verlag, April 1994.

....specified and subject to dynamic changes. Instead in this paper, we present hamlet, an incremental and inductive strategy learning system that acquires control knowledge for guiding the base level nonlinear planner prodigy4.0 [3] to efficiently achieve optimal solutions for complex problems [1, 2]. Although this paper describes the hamlet strategy learning tool in the context of prodigy, the main characteristics and contributions of this integration are general to any problem solver, in particular: ffl Definition of problem solving or planning as a decision making process where choice ....

....Inductive Learning module, which generalizes the learned rules by analyzing new examples of situations where the rules are applicable. We have devised methods for generalizing over the following aspects of the learned knowledge: state; subgoaling structure; interacting goals; and type hierarchy [2]. Example Consider that hamlet has previously learned the rule shown in Figure 4. The rule says that the planner should find a way of achieving (inside truck package1 truck1) before moving the carriers, plane1 and truck1 (even though flying an empty airplane will achieve one of the two planning ....

Daniel Borrajo and Manuela Veloso. Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94, pages 64--82. Springer Verlag, April 1994.

....hamlet learns control knowledge for individual decisions to guide the problem solver, by loosely explaining the trace of training episodes, and then refining the learned knowledge through experiencing positive and negative examples. We have described different aspects of hamlet s procedures in [1, 2]. hamlet is an ongoing research project. In this talk, we describe the multiple learning opportunities that hamlet addresses in the context of nonlinear planning, and the refinement of the learned knowledge. 2 Learning opportunities in nonlinear problem solving As a case study, we will present ....

Daniel Borrajo and Manuela Veloso. Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94. Springer Verlag, April 1994.

....applied to simple problem solving scenarios. Alternatively to these deductive domain theory dependent algorithms, inductive learning approaches incrementally acquire correct knowledge by observing a large set of 1 HAMLET stands for Heuristics Acquisition Method byLearning from sEarch Trees (Borrajo Veloso 1994; Veloso Borrajo 1994) 2 In addition to the type hierarchy, set of operators and inference rules that describe the primitive problem solving action model, a complete domain theory would include also a set of domain axioms that enables the proof of the universal truth of episodic ....

....scenarios. Alternatively to these deductive domain theory dependent algorithms, inductive learning approaches incrementally acquire correct knowledge by observing a large set of 1 HAMLET stands for Heuristics Acquisition Method byLearning from sEarch Trees (Borrajo Veloso 1994; Veloso Borrajo 1994). 2 In addition to the type hierarchy, set of operators and inference rules that describe the primitive problem solving action model, a complete domain theory would include also a set of domain axioms that enables the proof of the universal truth of episodic explanations. problem solving ....

Borrajo, D., and Veloso, M. 1994. Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94, 64--82. Springer Verlag.

....a positive example only if the decision leads to one of the globally best solutions and it was not the choice selected by the problem solving default heuristics. We implemented our learning approach in a system called hamlet, standing for Heuristics Acquisition Method by Learning from sEarch Trees (Borrajo and Veloso, 1994; Veloso and Borrajo, 1994) hamlet is integrated with prodigy4.0, the current nonlinear problem solver of the prodigy architecture for planning and learning (Carbonell et al. 1992) hamlet learns control knowledge incrementally and inductively to improve both the search efficiency of the ....

....leads to one of the globally best solutions and it was not the choice selected by the problem solving default heuristics. We implemented our learning approach in a system called hamlet, standing for Heuristics Acquisition Method by Learning from sEarch Trees (Borrajo and Veloso, 1994; Veloso and Borrajo, 1994). hamlet is integrated with prodigy4.0, the current nonlinear problem solver of the prodigy architecture for planning and learning (Carbonell et al. 1992) hamlet learns control knowledge incrementally and inductively to improve both the search efficiency of the problem solver and to improve ....

[Article contains additional citation context not shown here]

Borrajo, D. and Veloso, M. (1994). Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, pages 64--82, Sicily, Italy. Springer Verlag.

.... Problem Formulation Researchers in Machine Learning and in Planning have developed several systems that use simple planning problems to learn how to solve more difficult problems (among several others, Laird, Rosenbloom, Newell 1986; DeJong Mooney 1986; Minton 1988; Veloso Carbonell 1993; Borrajo Veloso 1994)) However, all current systems require that the simpler problems be provided by the user. This requirement is a gap in automated learning that we propose to fill. Using a previously untried approach, we are developing a system that will find auxiliary problems that are likely to be useful in ....

Borrajo, D., and Veloso, M. 1994. Incremental learning of control knowledge for nonlinear problem solving.

No context found.

Borrajo, D. and Veloso, M. (1994b). Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94, pages 64--82.

No context found.

Borrajo, D., & Veloso, M. (1994b). Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94, pp. 64--82 Springer Verlag.

No context found.

Borrajo, D., & Veloso, M. (1994b). Incremental learning of control knowledge for nonlinear problem solving. In Proceedings of the European Conference on Machine Learning, ECML-94, pp. 64--82 Springer Verlag.

No context found.

Borrajo, D., and Veloso, M. 1994a. Incremental learning of control knowledge for nonlinear problem solving.

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