Distributed parameter tuning for genetic algorithms

Genetic Algorithms (GA) is a family of search algorithms based on the mechanics of natural selection and biological evolution. They are able to efficiently exploit historical information in the evolution process to look for optimal solutions or approximate them for a given problem, achieving excellent performance in optimization problems that involve a large set of dependent variables. Despite the excellent results of GAs, their use may generate new problems. One of them is how to provide a good fitting in the usually large number of parameters that must be tuned to allow a good performance. This paper describes a new platform that is able to extract the Regular Expression that matches a set of examples, using a supervised learning and agent-based framework. In order to do that, GA-based agents decompose the GA execution in a distributed sequence of operations performed by them. The platform has been applied to Language induction problem, for that reason the experiments are focused on the extraction of the regular expression that matches a set of examples. Finally, the paper shows the efficiency of the proposed platform (in terms of fitness value) applied to three case studies: emails, phone numbers and URLs. Moreover, it is described how the codification of the alphabet affects to the performance of the platform.