SPTH
October 2010
Download PDF (370.31Kb) (You need to be registered on forum)First, three successful environments for artificial evolution in computer systems are analysed briefly. The organism in these enviroment are in a virtual machine with special chemistries. Two key-features are found to be very robust under mutations: Non-direct addressing and separation of instruction and argument.
In contrast, the x86 instruction set is very brittle under mutations, thus not able to achieve evolution directly. However, by making use of a special meta-language, these two key-features can be realized in a x86 system. This meta-language and its implementation is presented in chapter 2.
First experiments show very promising behaviour of the population. A statistically analyse of these population is done in chapter 3. One key-result has been found by comparison of the robustness of x86 instruction set and the meta-language: A statistical analyse of mutation densities shows that the meta-language is much more robust under mutations than the x86 instruction set.
In the end, some Open Questions are stated which should be addressed in further researches. An detailed explanation of how to run the experiment is given in the Appendix.