Steven Minton, Craig A. Knoblock, Daniel R. Kuokka, Yolanda Gil, Robert L. Joseph, and Jaime G. Carbonell. PRODIGY 2.0: The manual and tutorial. Technical Report CMUCS -89-146, School of Computer Science, Carnegie Mellon University,1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....a single metric rank. ranking function that takes a plan and returns a real number indicating metrically how good it is. Unfortunately, few estimators are known that are both, efficient and useful. A better idea is to use knowledge based search. Ideas first explored in SOAR [1] and then Prodigy [2]. ....

S. Minton, C. Knoblock, D. Koukka, Y. Gil, R. Joseph, and J. Carbonell. PRODIGY 2.0: The Manual and Tutorial. School of Computer Science, Carnegie Mellon University, Pittsburgh, 1989. CMU-CS-89146. 2

....of the resulting goals and assumptions can be predicted, however, the expansion has to take place right away. Control Knowledge Control knowledge in proof planning is used to reduce the search and to prefer proof plans with a structure that is comprehensible for the user. Several experiences (Minton 1989; Weld 1994) indicate the superiority of a separate representation of control knowledge by control rules. This modular representation is well suited for modi cations, for the user s comprehension, and for learning control knowledge. We adopted this approach. Currently, we distinguish the ....

....fertilize, symbolic evaluation that are important for a class of typical inductive proofs. Related from the knowledge acquisition perspective is the work of Bledsoe and Hines on special purpose theorem provers (Bledsoe Hines 1980) Related with respect to planning with control rules is Prodigy (Minton et al. 1989). ....

Minton, S.; Knoblock, C.; Koukka, D.; Gil, Y.; Joseph, R.; and Carbonell, J. 1989. PRODIGY 2.0: The Manual and Tutorial. School of Computer Science, Carnegie Mellon University, Pittsburgh. CMU-CS-89146.

....search via a miniature production system turned out to be a good idea. Usually, these rules refer to local decisions. They can, however, also express control knowledge referring to the global development of a plan. First, in SOAR [12] such a control was explored and in the Prodigy system [17] the ideas were refined. The experiences with a separate body of control rules in Prodigy are summarized in [19] The advantage of the factualcontrol knowledge distinction are modularity, reification of the control knowledge, selectivity in building learning modules, and compositionality of the ....

S. Minton, C. Knoblock, D. Koukka, Y. Gil, R. Joseph, and J. Carbonell. PRODIGY 2.0: The Manual and Tutorial. School of Computer Science, Carnegie Mellon University, Pittsburgh, 1989. CMU-CS-89-146.

....of control information by rules can be a basis for automatically learning control rules, as realized in some planning systems, e.g. in [31, 1, 13] Control rules are new in proof planning and in automated theorem proving. In problem solving planning, however, SOAR [12] and then Prodigy [32] were the first systems which used control rules. My experience and discussions at CMU and Edinburgh were very helpful in order to find and bring about this meta reasoning approach. 2.2 Integration of Constraint Solvers In many mathematical proofs, logical steps are naturally combined with some ....

S. Minton, C. Knoblock, D. Koukka, Y. Gil, R. Joseph, and J. Carbonell. PRODIGY 2.0: The Manual and Tutorial. School of Computer Science, Carnegie Mellon University, Pittsburgh, 1989. CMU-CS-89-146.

....search via a miniature production system turned out to be a good idea. Usually, these rules refer to local decisions. They can, however, also express control knowledge referring to the global development of a plan. First, in SOAR [14] such a control was explored and in the Prodigy system [18] the ideas were refined. Such a control is also described in [1] Meta level control rules can be found in Press [4] In Prodigy the control rules contain meta predicates that refer to the current state, the sequence of operators, etc. The experiences with a separate body of control rules in ....

S. Minton, C. Knoblock, D. Koukka, Y. Gil, R. Joseph, and J. Carbonell, PRODIGY 2.0: The Manual and Tutorial, School of Computer Science, Carnegie Mellon University, Pittsburgh, 1989. CMU-CS-89-146.

....strategy language and an interesting example. Here we summarize some recent developments along the same lines since that time. Maude, the basis for this work, is a logical language based on rewriting logic [24, 29, 26] It is therefore related to other rewriting logic languages such as Cafe [14], and ELAN [19] The equational language OBJ [17] can be regarded as a functional sublanguage of Maude. Maude is supported by a high performance execution engine built by Steven Eker and the rest of the Maude team [8, 7] This work supported by DARPA through Rome Laboratories Contract ....

....seems to be devoted to first order systems and, if I may be provocative, this is a curious state of affairs. First order logic is not expressive: consider how powerful resolution provers are, and how few applications they have. Adding set theory to first order logic yields an expressive system [14, 16], but formalizing concepts such as set comprehension, fx 2 A j P (x)g, and general union, S x2A B(x) requires a higher order syntax. Approaches based on first order syntax are not attractive we might have a clumsy language of combinators or be forced to define an auxiliary function every time ....

[Article contains additional citation context not shown here]

S. Minton, C. Knoblock, D. Koukka, Y. Gil, R. Joseph, and J. Carbonell. PRODIGY 2.0: The Manual and Tutorial. School of Computer Science, Carnegie Mellon University, Pittsburgh, 1989. CMU-CS-89-146.

....consists of (domain dependent) heuristics concerning decisions at choice points. They can be encoded into compiled procedures or as declarative control rules. Depending on the strictness of a heuristic, different types of control rules can be designed: choose, don t choose, and prefer rules [27]. For applying the state space refinement strategies other kinds of control knowledge can be used, as practiced, e.g. in Prodigy [27] This knowledge supports the decisions: ffl Choose goal ffl Choose bindings ffl Choose operator ffl Apply operator Corresponding to the decisions to be made in ....

....as declarative control rules. Depending on the strictness of a heuristic, different types of control rules can be designed: choose, don t choose, and prefer rules [27] For applying the state space refinement strategies other kinds of control knowledge can be used, as practiced, e.g. in Prodigy [27]. This knowledge supports the decisions: ffl Choose goal ffl Choose bindings ffl Choose operator ffl Apply operator Corresponding to the decisions to be made in the planning described in the previous sections, the additional classes of control knowledge need to be considered: ffl ....

[Article contains additional citation context not shown here]

S. Minton, C. Knoblock, D. Koukka, Y. Gil, R. Joseph, and J. Carbonell. PRODIGY 2.0: The Manual and Tutorial. School of Computer Science, Carnegie Mellon University, Pittsburgh, 1989. CMU-CS-89-146.

..... The ground step s is introduced into the plan , s a t is replaced by s OE t. For pre(s) fpg, the auxiliary constraint abs(g 2 ; p) is introduced. 4. 2 Control Knowledge in Proof Planning In OMEGA, the planning is controlled by the interpretation of control rules similar to those in Prodigy [14]. Control rules contain meta predicates that can describe the planning history, the state, resources, constraints. Currently, we dis Algorithm island refinement( Returns refinements of Parameters: introduce subproblem procedure Step selection: Pick abstract step sa for which abs(g) ....

S. Minton, C. Knoblock, D. Koukka, Y. Gil, R. Joseph, and J. Carbonell. PRODIGY 2.0: The Manual and Tutorial. School of Computer Science, Carnegie Mellon University, Pittsburgh, 1989. CMU-CS-89-146.

No context found.

Steven Minton, Craig A. Knoblock, Daniel R. Kuokka, Yolanda Gil, Robert L. Joseph, and Jaime G. Carbonell. PRODIGY 2.0: The manual and tutorial. Technical Report CMUCS -89-146, School of Computer Science, Carnegie Mellon University,1989.

....at the operator and at the goal ordering levels. Commitments are made during the search process, in contrast to a least commitment strategy [Sacerdoti, 1975, Tate, 1977, Wilkins, 1989] where decisions are deferred until all possible interactions are recognized. With the casualcommitment approach [Minton et al. 1989] , background knowledge, whether hand coded expertise, learned control rules, or heuristic evaluation functions, guides the efficient exploration of the most promising parts of the search space. Provably incorrect alternatives are eliminated and heuristically preferred ones are explored first. ....

Steven Minton, Craig A. Knoblock, Dan R. Kuokka, Yolanda Gil, Robert L. Joseph, and Jaime G. Carbonell. PRODIGY 2.0: The manual and tutorial. Technical Report CMUCS -89-146, School of Computer Science, Carnegie Mellon University, 1989.

....at the operator and at the goal ordering levels. Commitments are made during the search process, in contrast to a least commitment strategy [Sacerdoti, 1975, Tate, 1977, Wilkins, 1989] where decisions are deferred until all possible interactions are recognized. With the casualcommitment approach [Minton et al. 1989] , background knowledge, whether hand coded expertise, learned control rules, or heuristic evaluation functions, guides the efficient exploration of the most promising parts of the search space. Provably incorrect alternatives are eliminated and heuristically preferred ones are explored first. ....

Steven Minton, Craig A. Knoblock, Dan R. Kuokka, Yolanda Gil, Robert L. Joseph, and Jaime G. Carbonell. PRODIGY 2.0: The manual and tutorial. Technical Report CMUCS -89-146, School of Computer Science, Carnegie Mellon University, 1989.

....preferences. The work described here is a method for refining the specifications of operators, and it has been implemented in a version of the PRODIGY system augmented with capabilities for execution monitoring and dynamic replanning. 2. The Role of Experimentation in PRODIGY The PRODIGY system [Minton et al. 89, 89, Carbonell et al. 90] is a general purpose problem solver designed to provide an underlying basis for machine learning research. The appendix presents an overview of the basic architecture and the different learning mechanisms in the system. PRODIGY can improve its performance, by ....

Minton, S., Knoblock, C. A., Kuokka, D. R., Gil, Y., Joseph, R. L., Carbonell, J. G., PRODIGY 2.0: The Manual and Tutorial, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA, Technical Report CMU-CS-89-146, 1989.

....even for people. This paper identifies a criterion for selecting useful abstractions, describes a tractable algorithm for generating them, and empirically demonstrates that the abstractions reduce search. The abstraction learner, called alpine, is integrated with the prodigy problem solver [ Minton et al. 1989b, Carbonell et al. 1991 ] and has been tested on large problem sets in multiple domains. Introduction Hierarchical problem solving uses abstraction to reduce the complexity of search by dividing up a problem into smaller subproblems [ Korf, 1987, Knoblock, 1990 ] Given a problem space and a ....

....problem solver that employs a state space representation. Results alpine produces useful abstraction hierarchies in a number of problem domains. This section demonstrates the effectiveness of alpine s abstractions in a machine shop scheduling domain and a robot planning domain [ Minton, 1988, Minton et al. 1989a ] These domains were originally used to evaluate explanationbased learning (ebl) in prodigy. A problem in the machine shop scheduling domain involves finding a valid sequence of machining operations and scheduling the operations to produce various parts. The robot planning domain is an ....

Steven Minton, Craig A. Knoblock, Daniel R. Kuokka, Yolanda Gil, Robert L. Joseph, and Jaime G. Carbonell. PRODIGY 2.0: The manual and tutorial. Technical Report CMUCS -89-146, School of Computer Science, Carnegie Mellon University, 1989.

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