Created by W.Langdon from gp-bibliography.bib Revision:1.8010
Two techniques implementing those methods are described in this work. The first one, named SAGE, extends the sampling-based strategy underlying evolutionary algorithms to perform search in trees and directed acyclic graphs. The second technique considers coevolutionary learning, a paradigm which involves the embedding of adaptive agents in a fitness environment that dynamically responds to their progress. Coevolution is proposed as a framework in which evolving agents would be permanently challenged, eventually resulting in continuous improvement of their performance. After identifying obstacles to continuous progress, the concept of an ``Ideal'' trainer is presented as a paradigm which successfully achieves that goal by maintaining a pressure toward adaptability.
The different algorithms discussed in this dissertation have been applied to a variety of difficult problems in learning and combinatorial optimization. Some significant achievements that resulted from those experiments concern: (1) the discovery of new constructions for 13-input sorting networks using fewer comparators than the best known upper bound, (2) an improved procedure for the induction of DFAs from sparse training data which ended up as a co-winner in a grammar inference competition, and (3) the discovery of new cellular automata rules to implement the majority classification task which outperform the best known rules.
By describing evolutionary algorithms from the perspective of statistical inference techniques, this research work contributes to a better understanding of the underlying search strategies embedded in EC techniques. In particular, an extensive analysis of the coevolutionary paradigm identifies two fundamental requirements for achieving continuous progress. Search and machine learning are two fields that are closely related. This dissertation emphasises this relationship and demonstrates the relevance of the issue of generalisation in the context of coevolutionary races.",
GP, eg starting on page 169",
Genetic Programming entries for Hugues Juille