Translation Tables: A Genetic Code in a Evolutionary Algorithm

The genetic code that maps triples of DNA onto amino acids, is a central part of the biochemistry of life. In this study we incorporate an analogous code, called a translation table, into the self-avoiding walk test problem. Use of a translation table permits evolution of both the distribution of commands and the behavior of the mutation operator. It thus can evolve to encode two types of domain knowledge about the test problem. The translation tables are shown to specialize to specific cases of the test problem but yield no significant improvement in performance. The emergence of encoded problem-specific knowledge in the translation tables is demonstrated. A translation table constructed from extrapolation of the evolutionary trend yields a performance improvement, suggesting that the current algorithm would require more time than that allocated in the experiments to locate translation tables that would enhance performance. A tentative technique for overcoming this limitation is outlined.