generalfactbullishThe discrete training problem of shallow neural networks can be replaced with a well-posed continuum variational surrogate using λ-convex functionalsMachine Learning (Statistics)27 Jul 2026http://arxiv.org/abs/2607.02003v1