Efficient Evolution of High Entropy RNGs Using Single Node Genetic Programming

摘要: Random Number Generators are an important aspect of many modern day software systems, cryptographic protocols and modelling techniques. To be more accurate, it is Pseudo (PRNGs) that commonly used over their expensive, less practical hardware based counterparts. Given PRNGs rely on some deterministic algorithm (typically a Linear Congruential Generator) we can leverage Shannon's theory information as our fitness function in order to generate these algorithms by evolutionary means. In this paper compare traditional Genetic Programming (GP) against its graph implementation, Single Node (SNGP), for task. We show with SNGPs unique program structure use dynamic programming, possible obtain smaller, higher entropy PRNGs, six times faster produced at solution rate twice achieved using Koza's standard GP model. also the obtained from methods produce outputs than other widely Hardware RNGs (specifically recordings atmospheric noise), well surpassing them variety statistical tests presented NIST RNG test suite.