generalfactneutralWhen learning improves steadily with more data, deeper memory in generative selection never hurts performanceNeural and Evolutionary Computing01 Aug 2026http://arxiv.org/abs/2607.22211v1