← back to paper
arxiv: 2605.18528 · 2 revisions
Scale-Invariant Neural Network Optimization: Norm Geometry and Heavy-Tailed Noise