Lionel Martin

CoFKM: A Centralized Method for Multiple-View Clustering

This paper deals with clustering for multi-view data, i.e. objects described by several sets of variables or proximity matrices. Many important domains or applications such as Information Retrieval, biology, chemistry and marketing are concerned by this problematic. The aim of this data mining research field is to search for clustering patterns that perform a consensus between the patterns from different views. This requires to merge information from each view by performing a fusion process that identifies the agreement between the views and solves the conflicts. Various fusion strategies can be applied, occurring either before, after or during the clustering process. We draw our inspiration from the existing algorithms based on a centralized strategy. We propose a fuzzy clustering approach that generalizes the three fusion strategies and outperforms the main existing multi-view clustering algorithm both on synthetic and real datasets.

On Learning Constraint Problems

It is well known that modeling with constraints networks require a fair expertise. Thus tools able to automatically generate such networks have gained a major interest. The major contribution of this paper is to set a new framework based on Inductive Logic Programming able to build a constraint model from solutions and non-solutions of related problems. The model is expressed in a middle-level modeling language. On this particular relational learning problem, traditional top-down search methods fall into blind search and bottom-up search methods produce too expensive coverage tests. Recent works in Inductive Logic Programming about phase transition and crossing plateau shows that no general solution can face all these difficulties. In this context, we have designed an algorithm combining the major qualities of these two types of search techniques. We present experimental results on some benchmarks ranging from puzzles to scheduling problems.

Indexical-Based Solver Learning

The pioneering works of Apt and Monfroy, and Abdennadher and Rigotti have shown that the construction of rule-based solvers can be automated using machine learning techniques. Both works implement the solver as a set of CHRs. But many solvers use the more specialized chaotic iteration of operators as operational semantics and not CHRs rewriting semantics. In this paper, we first define a language-independent framework for operator learning and then we apply it to the learning of partial arc-consistency operators for a subset of the indexical language of Gnu-Prolog and show the effectiveness of our approach by two implementations. On tested examples, Gnu-Prolog solvers are learned from their original constraints and powerful propagators are found for user-defined constraints.

Disjunctive Learning with a Soft-Clustering Method

In the case of concept learning from positive and negative examples, it is rarely possible to find a unique discriminating conjunctive rule; in most cases, a disjunctive description is needed. This problem, known as disjunctive learning, is mainly solved by greedy methods, iteratively adding rules until all positive examples are covered. Each rule is determined by discriminating properties, where the discriminating power is computed from the learning set. Each rule defines a subconcept of concept to be learned with these methods. The final set of sub-concepts is then highly dependent from both the learning set and the learning method.

Induction of Constraint Logic Programs

Inductive Logic Programming is mainly concerned with the problem of learning concept definitions from positive and negative examples of these concepts and background knowledge. Because of complexity problems, the underlying first order language is often restricted to variables, predicates and constants. In this paper, we propose a new approach for learning logic programs containing function symbols other than constants. The underlying idea is to consider a domain that enables to interpret the function symbols and to compute the interest of a given value for discriminating positive and negative examples. This is modelized in the framework of Constraint Logic Programming and the algorithm that we propose enables to learn some constraint logic programs. This algorithm has been implemented in the system ICC. In order to reduce the complexity, biases have been introduced, as for instance the form of constraints that can be learned, the depth of a term or the size of the constraints.

Integrating Pairwise Constraints into Clustering Algorithms: Optimization-Based Approaches

In this paper we introduce new models for semi-supervised clustering problem, in particular we address this problem from the representation space point of view. Given a dataset enhanced with constraints (typically must-link and cannot-link constraints) and any clustering algorithm, the proposed approach aims at learning a projection space for the dataset that satisfies not only the constraints but also the required objective of the clustering algorithm on unenhanced data. We propose a boosting framework to weight the constraints and infers successive projection spaces in such a way that algorithm performance is improved. We experiment this approach on standard UCI datasets and show the effectiveness of our algorithm.

A Relational Data Mining Tool Based On Genetic Programming

In this paper, we present a Data Mining tool based on Genetic Programming which enables to analyze complex databases, involving several relation schemes. In our approach, trees represent expressions of relational algebra and they are evaluated according to the way they discriminate positive and negative examples of the target concept. Nevertheless, relational algebra expressions are strongly typed and classical genetic operators, such as mutation and crossover, have been modified to prevent from building illegal expressions. The Genetic Programming approach that we have developed has been modeled in the framework of constraints.

Learning Linear Constraints in Inductive Logic Programming

In this paper, we present a system, called ICC, that learns constrained logic programs containing function symbols. The particularity of our approach is to consider, as in the field of Constraint Logic Programming, a specific computation domain and to handle terms by taking into account their values in this domain. Nevertheless, an earlier version of our system was only able to learn constraints Xi=t, where Xi is a variable and t is a term. We propose here a method for learning linear constraints. It has already been a lot studied in the field of Statistical Learning Theory and for learning Oblic Decision Trees. As far as we know, the originality of our approach is to rely on a Linear Programming solver. Moreover, integrating it in ICC enables to learn non linear constraints.

A Language-Based Similarity Measure

This paper presents an unified framework for the definition of similarity measures for various formalisms (attribute-value, first order logic...). The underlying idea is that the similarity between two objects does not depend only on the attribute values of the objects, but more especially on the set of the potentially relevant definitions of concepts for the problem considered. In our framework, the user defines a language with a grammar to specify the similarity measure. Each term of the language represents a property of the objects. The similarity between two objects is the probability that these two objects both satisfy or both reject simultaneously the properties of the given language. When this probability is not computable, we use a stochastic generation procedure to approximate it. This measure can be applied for both clustering and classification tasks. The empirical evaluation on common classification problems shows a very good accuracy.

Interactive and progressive constraint definition for dimensionality reduction and visualization

Projecting and visualizing objects in a two- or tree-dimension space is a standard data analysis task. In addition to this visualization it might be of interest to allow the user to add knowledge in the form of (di)similarity constraints among objects, when those appear either too close or too far in the observation space. In this paper we propose three kinds of constraints and present a resolution method that derives from PCA. Experiments have been performed with both synthetic and usual datasets. They show that a relevant representation can be achieved with a limited set of constraints.