I just got back from two congresses. The first one, the International Conference on Evolutionary Computation (ICEC) was held in Albufera, Valencia (Spain) from 24 to 26 of November, as a part of the 2nd International Joint Conference on Computational Intelligence (IJCCI 2010), together with the Int. Conf. on Fuzzy Computation (ICFC) and the Int. Conf. on Neural Computation (ICNC). At ICEC I presented the paper “Investigating Replacement Strategies for the Adaptive Dissortative Mating Genetic Algorithm” (Fernandes, Merelo and Rosa, 2010), which addresses one of the open questions of my PhD thesis. The abstract:
This paper investigates the effects of modifying the Adaptive Dissortative Mating Genetic Algorithm (ADMGA) replacement strategy on the performance of the algorithm in dynamic problems. ADMGA is a variation of the standard GA with a mating restriction based on the genotypic similarity of the individuals. Dissimilar individuals mate more often than expected by chance and, as a result, genetic diversity throughout the run is maintained at a higher level. ADMGA was previously tested in dynamic optimization problems with promising results: the algorithm shows to outperform standard GAs and state-of-the-art approaches on several problems and dynamics. However, the performance of the algorithm degrades when the frequency of changes increases. Due to the premises under which ADMGA was tested, it has been argued that the replacement strategy that emerges from the algorithm’s dissortative mating strategy may be harming the performance in such situations. This study proposes alternative replacement schemes with the objective of improving ADMGA’s performance on fast changing environments (without damaging the performance on slower ones). The strategies maintain the simplicity of the algorithm, i.e., the parameter set is not increased. The replacement schemes were tested in dynamic environments based on stationary functions with different characteristics, showing to improve standard ADMGA’s performance in fast dynamic problems.
Then I headed for Djerba, Tunisia, to attend the International Conference on Metaheuristics and Nature Inspired Computing (META 2010), where first I presented the preliminary results of one our teams’ most motivating research projects, summarized in the paper “Towards an Automatic Sleep Classification Procedure Using Wavelet Based Feature Extraction” (Mora, Fernandes, Herrera, Castillo, Rosa and Merelo, 2010). In this project, we are using several algorithms — such as Support Vector Machines, G-Prop, K-nn and “our” clustering and classification algorithm, KANTS — to try to solve the extremely difficult problem of automatic sleep stages classification. A more detailed reported will appear soon.
Finally, and still in META, I presented the paper “Optimizing Evolutionary Algorithms at Program Level” (Merelo, Mora, Castillo, Laredo and Fernandes, 2010), an attempt to define good programming practices for Evolutionary Algorithms and identify the most relevant bottlenecks that may appear in this type of code. I leave you the slide presentation of this last paper: