Para-Gaussian Approximations of Random Deep Neural Networks
Slightly non-Gaussian approximation for random deep networks, where Gaussian limits degrade with depth.
Explainer · on the wire
The explainer for this story is still on the wire.
The claim, the mechanism, and the figures land here once generated.
Source · arXiv wire
arxiv.org/abs/2610.37388cs.LG · math.PR