{"id":"arxiv/2605.02108","url":"https://arxiv.org/abs/2605.02108","tags":["cs.LG","math.DG"],"score":0,"title":"Geometric and Spectral Alignment for Deep Neural Network I","domain":"cs.LG","source":"arxiv","authors":["Ziran Liu","Wei Wang","Jinhao Wang","Pengcheng Wang","Xinyi Sui","Cihan Ruan","Nam Ling","Wei Jiang"],"reviews":0,"subject":"cs.LG","summary":"Deep residual architectures are modeled as products of near-identity Jacobians. This paper proves deterministic quotient-geometric estimates for singular spectra of Frobenius-normalized layer factors, emphasizing a normalized top-radial Cartan coordinate and fitted power-law chart. Full-rank factors are mapped from $\\mathrm{GL}(d)$ to the positive cone by $A\\mapsto A^\\top A$, then to ordered eigenvalue data. Under Frobenius normalization, exact power-law spectra form a trace-normalized Cartan orbit. This orbit is a Gibbs family on ranks, a Fisher information line, and a Bures--Wasserstein…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-05-04","reactions":{}}