generalfactneutralClosed-loop generative selection methods have lacked a rigorous convergence theory because retraining breaks the Markov propertyNeural and Evolutionary Computing01 Aug 2026http://arxiv.org/abs/2607.22211v1