Discriminative Markov Logic Network Structure Learning Based on Propositionalization and chi2%2DTest.

On Learning Constraint Problems.
In this thesis we consider reasoning with preferences in a number of different AI problems. We start with a specific area as, temporal reasoning, for which there is specialized reasoning machinery. In particular, we consider quantitative temporal constraints and we add preferences in such a context. We discuss the complexity for solving temporal constraint satisfaction problems with preferences in general, and we identify a tractable subclass: simple temporal problems with semi-convex preference functions. For this subclass, we consider finding optimal solutions with respect to the (fuzzy) maximin criterion and to the Pareto criterion. In the first case, we propose two solvers based on two different approaches. One enforces the notion of path consistency adapted to deal also with preferences. We show that enforcing such a property is sufficient to solve an STPP and it is a polynomial time algorithm. The second solver, reduces the problem to that of solving several hard constraint problems. We show that also this solver is polynomial. We have also designed a solver that finds Pareto optimal solutions. It applies one of the solvers for fuzzy-optimal solutions several times, to problems obtained from the original one by changing some constraints. Also this solver is polynomial. We have implemented and tested the solvers for fuzzy optimals. The tests, performed on randomly generated problems, show that the second solver is much faster, although it allows to represent less general preferences. We also consider the problem of actually obtaining the preferences on all the temporal constraints. Practically, it may not be feasible to retrieve all such preferences by a user. We, vii  thus, apply a machine learning technique, inductive learning, to overcome this problem. In partic...
A Comparison between Two Statistical Relational Models.

QuantMiner A Genetic Algorithm for Mining Quantitative Association Rules.
In this paper, we propose QUANTMINER, a mining quantitative association rules system. This system is based on a genetic algorithm that dynamically discovers “good ” intervals in association rules by optimizing both the support and the confidence. The experiments on real and artificial databases have shown the usefulness of QUANTMINER as an interactive data mining tool.
On the Discovery of Exception Rules A Survey.

A Contribution to the Use of Decision Diagrams for Loading and Mining Transaction Databases.

Une mesure de proximit%E9 et une m%E9thode de regroupement pour l%27aide %E0 l%27acquisition d%27ontologies sp%E9cialis%E9es.

Structuring Natural Language Data by Learning Rewriting Rules.

A Proximity Measure and a Clustering Method for Concept Extraction in an Ontology Building Perspective.

Extraction de r%E8gles d%27association quantitatives %2D Application %E0 des donn%E9es m%E9dicales.

Estimation of the Density of Datasets with Decision Diagrams.

PoBOC An Overlapping Clustering Algorithm
Abstract. In this paper, we propose an approach for constructing clusters of related terms that may be used for deriving formal conceptual structures in a later stage. In contrast to previous approaches in this direction, we explicitly take into account the fact that words can have different, possibly even unrelated, meanings. To account for such ambiguities in word meaning, we consider two alternative soft clustering techniques, namely Overlapping Pole-Based Clustering (PoBOC) and Clustering by Committees (CBC). These soft clustering algorithms are used to detect different contexts of the clustered words, resulting in possibly more than one cluster membership per word. We report on initial experiments conducted on textual data from the tourism domain. 1
PoBOC  un algorithme de %22soft%2Dclustering%22. Applications %E0 l%27apprentissage de r%E8gles et au traitement de donn%E9es textuelles.

BoolLoader  un chargeur efficace d%E9di%E9 aux bases de transactions denses.

DDOC Overlapping Clustering of Words for Document Classification.

Apprentissage de r%E8gles caract%E9ristiques.

Learning Characteristic Rules Relying on Quantified Paths.

Mining maximal frequent itemsets by a boolean approach.

A Genetic Algorithm for Propositionalization.

Composition of Biasses of Inductive Logic Programming.

Mining Relational Databases.

An Application of Association Rules Discovery to Geographic Information Systems.

Learning in Constraint Databases.

A Relational Data Mining Tool Based On Genetic Programming.

Learning Linear Constraints in Inductive Logic Programming.

Systematic Predicate Invention in Inductive Logic Programming.

Efficient Induction of Numerical Constraints.

Induction of Constraint Logic Programs.
. This paper provides a brief introduction and overview of the  emerging area of Inductive Constraint Logic Programming (ICLP). It  discusses some of the existing work in the area and presents some of the  research issues and open questions that need to be addressed.  1 Introduction  Inductive Logic Programming (ILP) refers to a class of machine learning algorithms where the agent learns a first-order theory from examples and background knowledge. The ILP framework in machine learning is perhaps the most general of all because of the complexity of the concepts learned. The use of first-order logic programs as the underlying representation makes ILP systems more powerful and useful than the conventional empirical machine learning systems. ILP systems have been successfully used in a variety of real life domains including mesh design, protein synthesis, games and fault diagnosis. Most of the first-order representations used in ILP systems are variants of horn-clause based clausal logic o...
Hierarchical Conceptual Clustering in a First Order Representation.

Learning Query Rules for Optimizing Databases with Update Rules.

Apprentissage de R%E8gles et Bases de Donn%E9es D%E9ductives.

Building a Tool for Software Code Analysis A Machine Learning Approach.

Inductive Learning of Normal Clauses.
this paper, we are interested in the induction of normal clauses. Several semantics have been proposed for normal programs, we consider here the well-founded semantics, based on a three-valued logics. We have generalized to this semantic the classical constraint: all the positive examples must be covered by the learned program and all the negative ones must be rejected. In the case of inductive learning, this constraint seems too strong and we have defined a weaker criteria: we require that a positive (resp. negative) example is not considered as F alse (resp. T rue) by the learned program. This study has been applied to the system FOIL. Some positive examples that were covered by the program during the learning process were not in the well-founded semantics. and we define biases to solve this problem.
Building a Tool for Software Code Analysis A Machine Learning Approach.

The Use of Analogy in Incremental SBL.

