Spectral Asymptotics of Neural Network Loss Landscapes: An Exact Decomposition of the Curvature Exponent

• URL: http://arxiv.org/abs/2606.02596v1
• Date: Fri, 22 May 2026 23:31:38 GMT
• Title: Spectral Asymptotics of Neural Network Loss Landscapes: An Exact Decomposition of the Curvature Exponent
• Abstract summary: We prove the Spectral Alignment Decomposition: $= dlog_k / dlog_k$, where $_k$ measures alignment between Kronecker factor eigenbases and gradient singular directions.<n>This reduces "why does $$ vary?" to a geometric question we answer for LayerNorm, residual connections, and softmax heads.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The curvature exponent $α$ in $h_k \propto σ_k^α$ -- governing how Hessian eigenvalues scale with gradient singular values -- varies systematically across layer types ($α\approx 2$ for convolutions, $\approx 1$ for transformer attention, $< 1$ for MLP up-projections). Why? We prove the Spectral Alignment Decomposition: $α= 2 + d\logΦ_k / d\logσ_k$, where $Φ_k$ measures alignment between Kronecker factor eigenbases and gradient singular directions. This reduces "why does $α$ vary?" to a geometric question we answer for LayerNorm, residual connections, and softmax heads. The decomposition implies a spectral transfer identity $s = αγ$ linking curvature exponent, effective gradient rank-decay $γ$, and Hessian decay exponent $s$. The identity is algebraic; its empirical content is that $α$ and $γ$, fit on independent data (HVPs vs. SVD), recover $s$ to ~2% median error across 93 layers, five architectures, and three datasets -- with no free parameters. A zeta-function bound on participation ratio shows curvature concentrates onto effectively one direction per layer. As a proof of concept, we derive the architecture-adaptive preconditioner $T(σ;α)$ and show that Spectral Newton -- implementing $T$ in the gradient singular basis -- outperforms AdamW on vision benchmarks where $α\approx 2$.

Related papers

• Trace-Norm Overlaps of Quantum States: Interpolation, Equality, and Data-Processing Rigidity [0.0]
  We show that the difference between the Petz overlap and the directed trace-norm overlap exactly when $$ and $$.<n>We also prove that the difference between the Petz overlap and the directed trace-norm overlap vanishes exactly when $$ and $$.
  arXiv  Detail & Related papers  (2026-09-13T00:28:33Z)
• Asymptotically Optimal Regret for Reinforcement Learning without Horizon Dependence [54.59847568544922]
  We study horizon-free regret for finite-horizon time-homogeneous Markov decision processes with $A$ states, $A$ actions, horizon $H$, and per-trajectory total reward bounded by $1$.<n>We prove a regret upper bound [tilde O(sqrtSAK+S8A3)] with failure probability $$, where $K$ is the number of episodes and $tilde O(sqrtSAK+S3K)$ hides $mathsfpoly
  arXiv  Detail & Related papers  (2026-07-22T07:42:19Z)
• Efficient Mean Curvature Computation on High-Dimensional Data Manifolds [52.452902154360565]
  Estimating local mean curvature at each point of a high-dimensional dataset is a key ingredient of machine learning algorithms.<n>This paper introduces two complementary contributions that together reduce this cost by several orders of magnitude.<n> Experiments on real-world datasets confirm speedups of 50 to 300 times relative to the original implementation.
  arXiv  Detail & Related papers  (2026-06-04T16:04:31Z)
• When Does $\ell_2$-Boosting Overfit Benignly? High-Dimensional Risk Asymptotics and the $\ell_1$ Implicit Bias [15.113649527486276]
  We show that benign overfitting fails at the linear rate.<n>We show that while this localization mechanism should persist in the presence of signals, the exact signal-noise decomposition remains an open problem.
  arXiv  Detail & Related papers  (2026-05-07T14:14:09Z)
• The Spectral Edge Thesis: A Mathematical Framework for Intra-Signal Phase Transitions in Neural Network Training [0.0]
  We show that phase transitions in neural network training are controlled by the spectral gap of the rolling-window Gram matrix of parameter updates.<n>Adiabatic parameter $mathcalA = |G|_F / (, g2)$ controls circuit stability: $mathcalA ll 1$ (plateau), $mathcalA sim 1$ (phase transition), $mathcalA gg 1$ (forgetting)
  arXiv  Detail & Related papers  (2026-03-30T20:10:22Z)
• Rényi exponent landscape of multipartite entanglement in free-fermion systems [51.56484100374058]
  We show that the Rényi tripartite information $I_3() exhibits a qualitatively $exclusion-dependent scaling at small Fermi momentum.<n>A direct consequence is a replica obstruction: $I_m(n)/I_m(1) sim zm-1 to 0$ for all integer $n geq 2$, so the leading von Neumann signal cannot be reconstructed from integer Rényi data.
  arXiv  Detail & Related papers  (2026-03-09T22:27:00Z)
• Closed-form $\ell_r$ norm scaling with data for overparameterized linear regression and diagonal linear networks under $\ell_p$ bias [0.0]
  We give a unified, high-probability characterization for the scaling of the family of parameter norms.<n>We then study linear networks trained by descent.
  arXiv  Detail & Related papers  (2025-09-25T13:59:22Z)
• Learning Orthogonal Multi-Index Models: A Fine-Grained Information Exponent Analysis [54.57279006229212]
  Information exponent has played an important role in predicting the sample complexity of online gradient descent.<n>In this work, we show that by considering both second- and higher-order terms, we can first learn the relevant space using the second-order terms.<n>The overall sample and complexity of online SGD is $tildeO( d PL-1 )$.
  arXiv  Detail & Related papers  (2024-10-13T00:14:08Z)
• Neural network learns low-dimensional polynomials with SGD near the information-theoretic limit [75.4661041626338]
  We study the problem of gradient descent learning of a single-index target function $f_*(boldsymbolx) = textstylesigma_*left(langleboldsymbolx,boldsymbolthetarangleright)$<n>We prove that a two-layer neural network optimized by an SGD-based algorithm learns $f_*$ with a complexity that is not governed by information exponents.
  arXiv  Detail & Related papers  (2024-06-03T17:56:58Z)
• A Unified Framework for Uniform Signal Recovery in Nonlinear Generative Compressed Sensing [68.80803866919123]
  Under nonlinear measurements, most prior results are non-uniform, i.e., they hold with high probability for a fixed $mathbfx*$ rather than for all $mathbfx*$ simultaneously. Our framework accommodates GCS with 1-bit/uniformly quantized observations and single index models as canonical examples. We also develop a concentration inequality that produces tighter bounds for product processes whose index sets have low metric entropy.
  arXiv  Detail & Related papers  (2023-09-25T17:54:19Z)
• High-dimensional Asymptotics of Feature Learning: How One Gradient Step Improves the Representation [89.21686761957383]
  We study the first gradient descent step on the first-layer parameters $boldsymbolW$ in a two-layer network. Our results demonstrate that even one step can lead to a considerable advantage over random features.
  arXiv  Detail & Related papers  (2022-05-03T12:09:59Z)
• Tree-Projected Gradient Descent for Estimating Gradient-Sparse Parameters on Graphs [10.846572437131872]
  We study estimation of a gradient-sparse parameter vector $boldsymboltheta* in mathbbRp$. We show that, under suitable restricted strong convexity and smoothness assumptions for the loss, the resulting estimator achieves the squared-error risk $fracs*n log (1+fracps*)$ up to a multiplicative constant that is independent of $G$.
  arXiv  Detail & Related papers  (2020-05-31T20:08:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.