The framework of my research activities for the last five years has been artificial evolution, in particular genetic algorithms. Artificial evolution, this recent research discipline, belongs to the wider domain of artificial intelligence. Indeed, it is with the adaptive systems point of view that genetic algorithms were developped twenty five years ago by Professor John Holland. Their use either for function optimization or learning give evidence of this.
The work presented in my thesis  is grounded on two key notions: dynamical behaviors of population and polymorphism. Study of dynamical aspects is mandatory if one is interested in genetic algorithms as adaptive systems. Polymorphism is related to their own specificity: the manipulation of a population.
My contributions rely on a dual architecture based on the introduction of a meta level in interpretation of individuals: this allows to revisit both population dynamics and polymorphism. The maintenance of stable and polymorphic populations in the framework of multiobjective optimization [2,3], and the rich and complex behaviors obtained in the study of the minimal model of dual genetic algorithm , are examples of this.
The way the meta level modifies the search space exploration [6,7], led to an analogy between duality and the neutral theory of evolution. This theory claims that evolution occurs most by neutral mutations, i.e. mutations with no effects on the fitness value. The dual architecture allows the introduction such neutral mutations in a arbitrary space .
Moreover, I studied how to apply artificial evolution to neurogenesis, that is the automatic generation of artificial neural networks [1,4]. In this context, I was particularly interested in the characterization of the fitness landscapes induced by artificial neurogenesis .
Keywords : Genetic Algorithms, Artificial Evolution,
Artificial Intelligence, Adaptive Systems, Complex Systems, Chaos, Multiobjective
Optimization, Population Dynamics, Polymorphism.
5-Genetic Algorithm for Artificial Neurogenesis
Clergue M., Collard Ph. (1998). In Proceedings of the IEEE International Conference on Evolutionary Computation, Anchorage, 1998, pages 411-416. (gzipped ps, 90ko)
6-Fitness Distance Correlation,
as statistical measure of Genetic Algorithm
Collard Ph., Gaspar A., Clergue M. and Escazut C. (1998). In Proceedings of the European Conference on Artificial Intelligence, Brighton, 1998, pages 650-654, John Wihley & Sons. (gzipped ps, 40ko)
7-Genetic Heuristics for Search Space Exploration
Clergue M. and Collard Ph.(1999). In Proceedings of the International Joint Conference on Artificial Intelligence, Stockolm, 1999, pages 1218-1223, Morgan Kaufmann. (gzipped ps, 75ko)
8-Genetic Algorithms : from Hegemony to Chaos
Collard Ph. and Clergue M. (2000). Submitted to Complex Systems, Complex Systems Publication. (gzipped ps, 1.4Mo)
9-Synthetic Neutrality for Artificial Evolution
Collard Ph., Clergue M. and Michael Defoin-Platel (1999). In Artificial Evolution 99. (gzipped ps, 240ko)
10-Population Dynamics and Polymorphism in Genetic Algorithms
Clergue M. (1999) Ph. D. Dissertation (in french).
11-Misleading Functions Designed from Alternation
Collard Ph., Clergue M. and Bonnin F.(2000). In Congress on Evolutionary Computation 2000. (gzipped ps, 120ko)