Evolvable characteristic-based pseudo random number generation

摘要

We have recently succeeded in developing genetic algorithm (GA)-based random number generators (RNG) by encoding generators in genomes that manifest themselves as combinations of basic mathematical operators and randomized seeds/parameters. The combination of operators and parameters are "optimized" by selecting the "most fit" algorithm as measured by our GA's objective function (a metrized version of the Federal Information Protection Standard 140-1 (FIPS) statistical tests). Moreover, offline testing (with Marsaglia's Diehard test suite) shows that our characteristic-based RNGs perform well.