How random generator quality impacts genetic algorithm performance
Mark M. Meysenburg, Dan Hoelting, Duane McElvain, James A. Foster · 2002
It has been shown that pseudo-random number generator (PRNG) choice can affect simple genetic algorithm (GA) performance. However, these performance impacts are nonintuitive; PRNGs of poor quality can drive GAs to superior performance, for certain problems. The same PRNGs cause worse performance for other problems. In this paper we present a plausible explanation for this phenomenon: PRNGs of poor quality cause higher Vose discrepancy values than do higher quality PRNGs. Higher Vose discrepancy values could then be manifest as GA performance differences, as GA populations move toward fixed points of the Vose heuristic far away from the expectation. 1