MMODELYST
Papers/A Training-Time Diagnostic for Generalization via the Log-Alignment Ratio
PAP

A Training-Time Diagnostic for Generalization via the Log-Alignment Ratio

May 27, 2026

arXiv
Abstract

We study the log-alignment ratio (LAR), a measure of parameter-activation alignment, introduced in parameterization theory. We reformulate it as the overlap between a weight spectrum $p$ of the normalized squared singular values of a matrix and an activation spectrum $q$ of the normalized squared projections of inputs onto its singular directions. We show that unembedding LAR tracks the transition between memorization and generalization in two different settings by capturing the spread of $p$ and $q$ during training. In grokking, LAR predicts the effective dimension of the learned function: $k \approx n^{2(1-\text{LAR})}$, where $n$ is the input dimension of the matrix. In 3B-parameter language model pre-training, its deviation from a non-overfitting baseline tracks the generalization gap, and its rate of decline increases as overfitting approaches. LAR is computable from quantities available during the forward pass with negligible computational overhead, and requires no held-out validation data.

Select text to highlight · click a highlight to remove · saved in this browser only
Authors
Ali Shehper, Ashish Vaswani
Your notes (browser-local)
saved
arXiv:2605.28975