#### DMCA

## Substructure discovery in the SUBDUE system (1994)

Venue: | In Proc. of the Workshop on Knowledge Discovery in Databases |

Citations: | 77 - 3 self |

### Citations

765 | Knowledge acquisition via incre-mental conceptual clustering
- Fisher
- 1987
(Show Context)
Citation Context ...asses of graphs. High-probability components of the model represent substructure common to the model's instances. The Labyrinth system [21] extends the Cobweb incremental conceptual clustering system =-=[5]-=- to form hierarchical concepts of the structured objects. The upper-level components of the structured-object hierarchy represent substructures common to the input objects. Conklin and Glasgow [2] hav... |

436 | Understanding Line Drawings of Scenes with Shadows.
- Waltz
- 1975
(Show Context)
Citation Context ...hat were used for our experiments. To apply Subdue to image data, we extract edge information from the image and construct a graph representing the scene. The vertex labels follow the Waltz labelings =-=[22]-=- of junctions of edges. An edge arc represents the edge of an object in the image, and a space arc links non-connecting objects together. Figure 3 shows the scene used for our experiments. The data fo... |

378 |
Learning structural descriptions from examples, in The psychology of computer vision,
- Winston
- 1975
(Show Context)
Citation Context ...ion of distribution models. Section 10 summarizes our results with Subdue and discusses directions for future work. 2 Related Work Although specific to the blocks-world domain, Winston's Arch program =-=[24]-=- discovers substructure in order to deepen the hierarchical description of a scene and to group objects into more general concepts. Levinson [11] developed a system for storing labeled graphs in which... |

360 |
Stochastic complexity in statistical inquiry
- Rissanen
- 1989
(Show Context)
Citation Context ... discover graph-based production rules that can be interpreted as substructures common to the input graphs [9]. 3 Encoding Graphs The minimum description length (MDL) principle introduced by Rissanen =-=[19]-=- states that the best theory to describe a set of data is that theory which minimizes the description length of the entire data set. The MDL principle has been used for decision tree induction [17], i... |

322 | Inferrng decision trees using the minimum description length principle.
- Quinlan, Rivest
- 1989
(Show Context)
Citation Context ...en [19] states that the best theory to describe a set of data is that theory which minimizes the description length of the entire data set. The MDL principle has been used for decision tree induction =-=[17]-=-, image processing [10, 14, 15], concept learning from relational data [4], and learning models of non-homogeneous engineering domains [18]. We demonstrate how the minimum description length principle... |

305 |
Learning from observation: conceptual clustering. In
- Michalski, Stepp
- 1983
(Show Context)
Citation Context ...put graph described by the substructure. The coverage rule is motivated from research in inductive learning and provides that concept descriptions describing more input examples are considered better =-=[12]-=-. Although the MDL principle measures the amount of structure, the coverage rule includes the relevance of this savings with respect to the size of the entire input graph. Coverage is defined as the n... |

271 |
AutoClass: Bayesian classification system.
- Cheeseman, Kelly, et al.
- 1988
(Show Context)
Citation Context ...ion of attribute values. Therefore, every value is distinct and unrelated to other values. Section 9 discusses approaches to integrating Subdue with non-structural discovery systems such as AutoClass =-=[1]-=- which provide for the inclusion of distribution models. Section 10 summarizes our results with Subdue and discusses directions for future work. 2 Related Work Although specific to the blocks-world do... |

270 | Constructing simple stable descriptions for image partitioning.
- Leclerc
- 1989
(Show Context)
Citation Context ... best theory to describe a set of data is that theory which minimizes the description length of the entire data set. The MDL principle has been used for decision tree induction [17], image processing =-=[10, 14, 15]-=-, concept learning from relational data [4], and learning models of non-homogeneous engineering domains [18]. We demonstrate how the minimum description length principle can be used to discover substr... |

258 |
Syntactic Pattern Recognition and Applications,
- Fu
- 1982
(Show Context)
Citation Context ...mpression afforded by the substructures and using a linear-time approximation to graph isomorphism. Many results in grammatical inference are applicable to constrained classes of graphs (e.g., trees) =-=[6, 13]-=-, and discover graph-based production rules that can be interpreted as substructures common to the input graphs [9]. 3 Encoding Graphs The minimum description length (MDL) principle introduced by Riss... |

199 | Substructure discovery using minimum description length and background knowledge.
- Cook, Holder
- 1994
(Show Context)
Citation Context ...red substructure concepts allow abstraction over detailed structure in the original data and provide new, relevant attributes for interpreting the data. We describe a new version of our Subdue system =-=[7, 8, 3]-=- that discovers interesting substructures in structural data based on the minimum description length principle and optional background knowledge. The next section describes work related to substructur... |

58 | Concept formation in structured domains.
- Thompson, Langley
- 1991
(Show Context)
Citation Context ... storing graphs using a probabilistic graph model to represent classes of graphs. High-probability components of the model represent substructure common to the model's instances. The Labyrinth system =-=[21]-=- extends the Cobweb incremental conceptual clustering system [5] to form hierarchical concepts of the structured objects. The upper-level components of the structured-object hierarchy represent substr... |

34 |
Some Experiments in Applying Inductive Inference Principles to Surface Reconstruction,” IJCAI,
- Pednault
- 1998
(Show Context)
Citation Context ... best theory to describe a set of data is that theory which minimizes the description length of the entire data set. The MDL principle has been used for decision tree induction [17], image processing =-=[10, 14, 15]-=-, concept learning from relational data [4], and learning models of non-homogeneous engineering domains [18]. We demonstrate how the minimum description length principle can be used to discover substr... |

32 |
Part Segmentation for Object Recognition
- Pentland
- 1989
(Show Context)
Citation Context ... best theory to describe a set of data is that theory which minimizes the description length of the entire data set. The MDL principle has been used for decision tree induction [17], image processing =-=[10, 14, 15]-=-, concept learning from relational data [4], and learning models of non-homogeneous engineering domains [18]. We demonstrate how the minimum description length principle can be used to discover substr... |

31 |
A Sell Organizing Retrieval System for Graphs.
- Levinson
- 1985
(Show Context)
Citation Context ...fic to the blocks-world domain, Winston's Arch program [24] discovers substructure in order to deepen the hierarchical description of a scene and to group objects into more general concepts. Levinson =-=[11]-=- developed a system for storing labeled graphs in which individual graphs are represented by the set of vertices in a universal graph. Subgraphs of the universal graph used by several individual graph... |

19 |
of organization in perceptual forms
- Laws
- 1938
(Show Context)
Citation Context ...nto the Subduessystem are compactness and coverage. The first rule, compactness, is a generalization of Wertheimer's Factor of Closure, which states that human attention is drawn to closed structures =-=[23]-=-. A closed substructure has at least as many edges as vertices, whereas a non-closed substructure has fewer edges than vertices [16]. Compactness is thus defined as the ratio of the number of edges in... |

18 |
Grammatical Inference Based on Hyperedge Replacement.
- Jeltsch, Kreowski
- 1991
(Show Context)
Citation Context ...mmatical inference are applicable to constrained classes of graphs (e.g., trees) [6, 13], and discover graph-based production rules that can be interpreted as substructures common to the input graphs =-=[9]-=-. 3 Encoding Graphs The minimum description length (MDL) principle introduced by Rissanen [19] states that the best theory to describe a set of data is that theory which minimizes the description leng... |

15 | Fuzzy substructure discovery.
- Holder, Cook, et al.
- 1992
(Show Context)
Citation Context ...red substructure concepts allow abstraction over detailed structure in the original data and provide new, relevant attributes for interpreting the data. We describe a new version of our Subdue system =-=[7, 8, 3]-=- that discovers interesting substructures in structural data based on the minimum description length principle and optional background knowledge. The next section describes work related to substructur... |

13 |
Structural Methods in Pattern Recognition
- Miclet
- 1986
(Show Context)
Citation Context ...mpression afforded by the substructures and using a linear-time approximation to graph isomorphism. Many results in grammatical inference are applicable to constrained classes of graphs (e.g., trees) =-=[6, 13]-=-, and discover graph-based production rules that can be interpreted as substructures common to the input graphs [9]. 3 Encoding Graphs The minimum description length (MDL) principle introduced by Riss... |

11 |
A minimal encoding approach to feature discovery
- Derthick
- 1991
(Show Context)
Citation Context ...y which minimizes the description length of the entire data set. The MDL principle has been used for decision tree induction [17], image processing [10, 14, 15], concept learning from relational data =-=[4]-=-, and learning models of non-homogeneous engineering domains [18]. We demonstrate how the minimum description length principle can be used to discover substructures in complex data. In particular, a s... |

11 |
Graph clustering and model learning by data compression
- Segen
- 1990
(Show Context)
Citation Context ...dual graphs are represented by the set of vertices in a universal graph. Subgraphs of the universal graph used by several individual graphs suggest common substructure in the individual graphs. Segen =-=[20]-=- describes a system for storing graphs using a probabilistic graph model to represent classes of graphs. High-probability components of the model represent substructure common to the model's instances... |

10 |
Learning engineering models with the minimum description length principle
- Rao, Lu
- 1992
(Show Context)
Citation Context ... The MDL principle has been used for decision tree induction [17], image processing [10, 14, 15], concept learning from relational data [4], and learning models of non-homogeneous engineering domains =-=[18]-=-. We demonstrate how the minimum description length principle can be used to discover substructures in complex data. In particular, a substructure is evaluated based on how well it can compress the en... |

10 |
Unifying learning methods by colored digraphs
- Yoshida, Motoda, et al.
- 1993
(Show Context)
Citation Context ...tem for constructing an image hierarchy, similar to that of Labyrinth, used for discovering common substructure in a set of images expressed in terms of a set of predefined relations. The CLiP system =-=[25]-=- for graph-based induction iteratively discovers patterns (substructures) in graphs by expanding and combining patterns discovered in previous iterations. CLiP addresses the computational complexity b... |

9 |
Discovery of spatial concepts in crystallographic databases
- Conklin, Fortier, et al.
- 1992
(Show Context)
Citation Context ...tem [5] to form hierarchical concepts of the structured objects. The upper-level components of the structured-object hierarchy represent substructures common to the input objects. Conklin and Glasgow =-=[2]-=- have developed the i-mem system for constructing an image hierarchy, similar to that of Labyrinth, used for discovering common substructure in a set of images expressed in terms of a set of predefine... |

6 | Discovery of inexact concepts from structural data
- Cook, Holder, et al.
- 1993
(Show Context)
Citation Context ...red substructure concepts allow abstraction over detailed structure in the original data and provide new, relevant attributes for interpreting the data. We describe a new version of our Subdue system =-=[7, 8, 3]-=- that discovers interesting substructures in structural data based on the minimum description length principle and optional background knowledge. The next section describes work related to substructur... |

3 |
Discrete Mathemetical Structures for
- Prather
- 1976
(Show Context)
Citation Context ...re, which states that human attention is drawn to closed structures [23]. A closed substructure has at least as many edges as vertices, whereas a non-closed substructure has fewer edges than vertices =-=[16]-=-. Compactness is thus defined as the ratio of the number of edges in the substructure to the number of vertices in the substructure. The second rule, coverage, measures the fraction of structure in th... |

1 |
Part segmentation for object reco~uition
- Pentland
- 1989
(Show Context)
Citation Context ...best theory to describe a set of data is that theory which minimizes the description length of the entire data set. The NIDL principle has been used for decision tree induction [17], image processing =-=[10, 14, 15]-=-, concept learning from relational data [4], and learning models of non-homogeneous engineering domains [18]. We demonstrate how the minimum description length principle can be used to discover substr... |

1 |
Discrete Mathemetical Structures for Computer Science. Houghton Mimn Company
- Prather
- 1976
(Show Context)
Citation Context ...n [19] states that the best theory to describe a set of data is that theory which minimizes the description length of the entire data set. The NIDL principle has been used for decision tree induction =-=[17]-=-, image processing [10, 14, 15], concept learning from relational data [4], and learning models of non-homogeneous engineering domains [18]. We demonstrate how the minimum description length principle... |