#### DMCA

## A Pattern-Weight Formulation of Search Knowledge (1994)

Venue: | Computational Intelligence |

Citations: | 5 - 3 self |

### Citations

3886 |
Adaptation in Natural and Artificial Systems. The
- Holland
- 1975
(Show Context)
Citation Context ...e-general-than. Pattern A is more general than pattern B, if for all states S, if B(S) is true, A(S) is true. What we call patterns have also been called features [43], equivalence classes and schema =-=[14] in the li-=-terature. We prefer the term "pattern" as they are most useful when they occur regularly. We will see that they are also related to what have been called abstract states and when coupled wit... |

888 | Data Structures and Algorithms. - Aho, Ullman, et al. - 1983 |

334 | Algorithmic graph theory. - GIBBONS - 1985 |

198 | Mind over machine. - Dreyfus, Dreyfus - 1986 |

117 |
Architecture of Systems Problem Solving,
- Klir
- 1985
(Show Context)
Citation Context ... be something like a cheese; it can be sliced in an infinite number of ways--and when one has chosen his own pattern of slicing, he finds that other['s] cuts fall at the wrong places. --Kenneth Burke =-=[17]-=- 1 Motivations Korf [19, 20, 21] has done an excellent job of classifying and describing the traditional forms of search knowledge. This was facilitated by the recognition that planning with perfect i... |

83 |
On the Eficient Implementation of Production Systems,
- Forgy
- 1979
(Show Context)
Citation Context ...in IS(Q) do P(x) := IP(x) [ fQg \Gamma IP(Q) here. Other applications are possible [34]. For simply represented states and patterns such as tile-puzzle states even more efficient schemes are possible =-=[9, 12]-=-; the same is true for a set of structured pattern-weights that are fixed since heavy precompilation can enable the structure screening and comparison operations to be done simultaneously [35]. 7 Exte... |

82 | Thought and choice - Groot - 1978 |

79 |
The B* tree search algorithm: A best-first proof procedure.
- Berliner
- 1979
(Show Context)
Citation Context ... considered yet. Still there is room for research progress here. Perhaps, the weights associated with pws should be ranges (lower and upper bounds) and search algorithms built up around them as in B* =-=[4]-=-. These ranges associated with patterns provide a useful mechanism of encoding only partial knowledge of a domain. With subgoals and macro-operators more preciseness is required. 8 Conclusions and Ong... |

69 | The behavior of adaptive systems which employ genetic and correlation algorithms, - Bagley - 1976 |

60 |
Learning to Solve Problems by Searching for Macro-Operators.
- Korf
- 1985
(Show Context)
Citation Context ...ese; it can be sliced in an infinite number of ways--and when one has chosen his own pattern of slicing, he finds that other['s] cuts fall at the wrong places. --Kenneth Burke [17] 1 Motivations Korf =-=[19, 20, 21]-=- has done an excellent job of classifying and describing the traditional forms of search knowledge. This was facilitated by the recognition that planning with perfect information can be viewed as heur... |

25 |
Notes on the 15 puzzle
- Johnson
(Show Context)
Citation Context ...those that traditionally go with A* and other search algorithms are optimistic (lower bounds). Examples in this paper will be taken from the n \Theta n tile puzzles. The traditional small tile puzzle =-=[15], called a-=-n "eight puzzle," is a 3x3 matrix of squares. Eight of the squares are occupied by sliding tiles which are numbered 1-8. The ninth square is empty, allowing the other tiles to be moved. The ... |

22 | A unified theory of heuristic evaluation functions and its application to learning - Christensen, Korf - 1986 |

17 |
A theory of abstraction for hierarchical planning
- Knoblock
- 1990
(Show Context)
Citation Context ...ictories 4.60 67.40 28.00 Table 2.3: Cost/benefit comparison (with admissable size-2 heuristic) A planning strategy that we shall only touch on here is the use of abstraction in hierarchical planning =-=[19, 18]. In a com-=-mon use of this strategy, certain states are collected together to form "abstract states." Interestingly, here too, a direct mapping to the patternweight formulation is possible. As there is... |

2 |
and machine: chess achievements and chess thinking
- Hearst, Man
- 1977
(Show Context)
Citation Context ... is possible using a shallow search depth. [27, 28, 30, 32, 33]. It is our hope that this pattern-oriented, low-search approach will be more consistent with cognitive models of human chess performance=-=[13]-=-. Acknowledgements Steve Grimm wrote the initial program code and improved the descriptions of the algorithms. Richard Snyder and Ira Pohl helped with an early draft. Sherri Shahrokhi wrote the initia... |

2 |
An analysis of conjunctive-goal planning
- Joslin, David, et al.
- 1986
(Show Context)
Citation Context ...not necessarily partial-descriptions of a goal-state [19]. Algorithms based only on ordering partial goal states, perform no better than brute force in the worst case - additional knowledge is needed =-=[16]-=-. 8 5. Patterns The difference between patterns and subgoals is that subgoals are totally-ordered in the order in which they are to be achieved. It is this ordering information that give subgoals much... |

1 |
Developments with GPS
- Banerji
- 1988
(Show Context)
Citation Context ...pulation. 4.1 Weaknesses in the Macro-table Technique Despite the obvious strengths of the macro-table technique in terms of space, correctness and execution time efficiency there are some weaknesses =-=[3]-=-. The most serious weakness is that the solution paths produced by macro-tables are often far from optimal (as depicted in Table 4.2). Tables 4.2 through Table 5.3 are a running example based on state... |

1 |
Partial order maintenance: a tutorial
- Colin, Levinson
- 1990
(Show Context)
Citation Context ...e selection of the next macro. However, state evaluation can be done efficiently by using an associative retrieval algorithm that takes advantage of a partial ordering of patterns by more-generalthan =-=[6, 10, 24, 25, 34]. Suppose -=-for example that pattern R is known to be a generalization of pattern S. Now once we determine that R is a specialization of a "query" pattern Q, we know that S is also without performing fu... |

1 |
An automatically compilable network recognition network for structured patterns
- Hayes-Roth, Mostow
- 1975
(Show Context)
Citation Context ...in IS(Q) do P(x) := IP(x) [ fQg \Gamma IP(Q) here. Other applications are possible [34]. For simply represented states and patterns such as tile-puzzle states even more efficient schemes are possible =-=[9, 12]-=-; the same is true for a set of structured pattern-weights that are fixed since heavy precompilation can enable the structure screening and comparison operations to be done simultaneously [35]. 7 Exte... |