Searching for the optimal coding in genetic algorithms

A genetic algorithm´s (GA´s) convergence depends on the quality of the building blocks present in the population (building block hypothesis). We formally define the building blocks and the coding function, and we demonstrate that building blocks present in a population depend on the used coding function; by assuming the building block hypothesis to be true, the convergence of GAs depends on the used coding function. We present a method which finds a coding function allowing meaningful building blocks to be obtained and improving GA convergence. We give a quantitative criterion to measure the `quality´ of each coding function: the research for the optimal coding is formalized as a minimization problem which is solved through GAs. We present some examples which demonstrate the improvement in GA convergence obtained through the coding resulting from the use of the method described in the paper