Fast Escape, Slow Convergence: Learning Dynamics of Phase Retrieval under Power-Law Data

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Braun, Guillaume, Loureiro, Bruno, Minh, Ha Quang, Imaizumi, Masaaki
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908671220383744
author Braun, Guillaume
Loureiro, Bruno
Minh, Ha Quang
Imaizumi, Masaaki
author_facet Braun, Guillaume
Loureiro, Bruno
Minh, Ha Quang
Imaizumi, Masaaki
contents Scaling laws describe how learning performance improves with data, compute, or training time, and have become a central theme in modern deep learning. We study this phenomenon in a canonical nonlinear model: phase retrieval with anisotropic Gaussian inputs whose covariance spectrum follows a power law. Unlike the isotropic case, where dynamics collapse to a two-dimensional system, anisotropy yields a qualitatively new regime in which an infinite hierarchy of coupled equations governs the evolution of the summary statistics. We develop a tractable reduction that reveals a three-phase trajectory: (i) fast escape from low alignment, (ii) slow convergence of the summary statistics, and (iii) spectral-tail learning in low-variance directions. From this decomposition, we derive explicit scaling laws for the mean-squared error, showing how spectral decay dictates convergence times and error curves. Experiments confirm the predicted phases and exponents. These results provide the first rigorous characterization of scaling laws in nonlinear regression with anisotropic data, highlighting how anisotropy reshapes learning dynamics.
format Preprint
id arxiv_https___arxiv_org_abs_2511_18661
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Fast Escape, Slow Convergence: Learning Dynamics of Phase Retrieval under Power-Law Data
Braun, Guillaume
Loureiro, Bruno
Minh, Ha Quang
Imaizumi, Masaaki
Machine Learning
Scaling laws describe how learning performance improves with data, compute, or training time, and have become a central theme in modern deep learning. We study this phenomenon in a canonical nonlinear model: phase retrieval with anisotropic Gaussian inputs whose covariance spectrum follows a power law. Unlike the isotropic case, where dynamics collapse to a two-dimensional system, anisotropy yields a qualitatively new regime in which an infinite hierarchy of coupled equations governs the evolution of the summary statistics. We develop a tractable reduction that reveals a three-phase trajectory: (i) fast escape from low alignment, (ii) slow convergence of the summary statistics, and (iii) spectral-tail learning in low-variance directions. From this decomposition, we derive explicit scaling laws for the mean-squared error, showing how spectral decay dictates convergence times and error curves. Experiments confirm the predicted phases and exponents. These results provide the first rigorous characterization of scaling laws in nonlinear regression with anisotropic data, highlighting how anisotropy reshapes learning dynamics.
title Fast Escape, Slow Convergence: Learning Dynamics of Phase Retrieval under Power-Law Data
topic Machine Learning
url https://arxiv.org/abs/2511.18661