ICML-98 Submission #103

Structural Machine Learning with Galois Lattice and Graphs

Michel Liquiere
LIRMM, 161 Rue Ada, 34392 Montpellier Cedex  5
France

Jean Sallantin
LIRMM, 161 Rue Ada, 34392 Montpellier Cedex  5
France

Abstract

The main objective of this paper is to define a formal approach to
learning from examples, when examples are described by labelled
graphs. For this purpose, this paper use a formal model based on the
use of lattice theory and, more precisely, in the use of Galois
connection. The advantage of this formalization is that we can now use
Galois lattice model with structural description of the examples and
concepts in order to enlarge the domain of formal conceptual analysis.
An implementation follows called Graal (for GRAph And Learning). This
tool constructs a Galois Lattice for any description language, if two
operations (compare and generalisation) are defined for that
language. We have defined these operations for labelled graphs.

Keywords:
Galois Lattice, labelled Graphs, Graphs product, Generalization

Email:	liquiere@lirmm.fr

Phone:	04 67 41 85 85