Research Topics

Project:

• Multiobjective Feature selection with GA: The study of the sensitivity and the specificity of a classification (diagnostic) test constitute a powerful analysis since it provides to the specialists more detailed information. In this work, we propose the use of a multiobjective genetic algorithm for gene selection of microarray datasets, that performs the selection from the point of view of the sensitivity and the specificity.
• Protein docking with optimization methods: in the molecular docking problem several aspects are important to be considered. Hence, we propose to adopt a new tri-objective modeling wich will combine two energetic criteria and a surface criterion. This model will be implemented, thanks to the ParadisEO platform, in a multi-objectif genetic algorithm making a full flexible molecular docking. This algorithm will be included in the Docking@Grid software.
• Protein identification: during the design of a de novo protein sequencing method for protein identification, we have proposed a new function to compare proteins based on the Fast Fourier Transform proposed by A.L. Rockwood to compute isotopic distributions. Thanks to this algorithm it is possible from a protein in FASTA format to generate the corresponding theoretical spectrum. In order to measure the quality of this function, we have designed a protein identification engine based of it: ASCQ_ME (See softwares below).
• Solving problem under uncertainty: a large part of real-world optimization problems are also subject to uncertainties due to, e.g., noisy or approximated objective function(s), varying parameters or dynamic environments. Although evolutionary algorithms are commonly used to solve multi-objective optimization problems on the one hand and stochastic optimization problems on the other hand, very few approaches combine these two aspects simultaneously. This is a new challenge for evolutionary algorithms.

A prelimary study on a bi-objective flow-shop scheduling problem with stochastic processing times has been realised in Liefooghe at EMO7, and interesting perspectives have been designed. In particular, a fundamental aspect is the performance assessment of optimizers dealing with uncertainty.

• Dynamic optimization: the importance of dynamic optimization for real-life applications is generally accepted and represents an important challenge. In dynamic optimization problems, a small perturbation of the data during the search occurs. The data may concern the objective function or the constraints associated to the problem. An example of dynamic problems is mobile robot path planning in an environment with moving obstacles.

Our objective will be to study the adaptation of metaheuristics in solving dynamic optimization problems. For example, it seems that metaheuristics based on an adaptive memory such as evolutionary algorithms and swarm intelligence algorithms are well adapted to such problems. Indeed, optimization methods must be able to adapt under a continuous change of the environment by reusing information obtained during the search. For example, the generation of diverse solutions in the search space will facilitate the adaptation of solutions. Moreover, a parallel design and implementation of the proposed metaheuristics will be considered.

New: Research a candidate for a phd thesis and for a postdoc: please contact me !

Softwares:

Thanks you to the students that contribute to these tools.

• Rule Mining

     --> ASGARD is a genetic rule mining algorithm developped in C

     --> Association Rules viewer is a sotfware developped in Java which is able to give a visual representation of association Rules.

• Multiobjective Optimization

• Protein indentification

• Docking

Phd Thesis:

Abstract

Firstly, we give a state of art of metaheuristics for knowledge discovery and in particular the use of genetic algorithms. We are particular interested in three fondamental aspects of metaheuristics: the representation of a solution, the fitness function and the choice of operators. Then, we introduce two cases of study that come from a collaboration with the Institute of Biology of Lille about the search for predisposition genetic factors for some multifactorial diseases (diabetes, obesity). We propose to model these problems as knowledge discovery tasks. Then, we modele the identified knowledge discovery tasks as optimization problems and we propose genetic algorithm scheme which has several advanced diversification and intensification mechanisms tasks. These mechanisms are tested separately to evaluate their improvements. We also integrate some knowledge of the biological domain in order to solve the problems under study. This integration is made while designing fitness functions and proposed mechanisms. Finally, some parallelism models are used.

Key words: Combinatorial Optimization, Genetic Algorithms, Genomic, Multifactorial disease, Knowledge discovery, Adaptive operators

Texte intégral disponible en français pdf ou ps(.zip de 2.1 Mo)