The high-dimensional asymptotics of first order methods with random data

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Celentano, Michael, Cheng, Chen, Montanari, Andrea
Format: Preprint
Published: 2021
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918452660273152
author Celentano, Michael
Cheng, Chen
Montanari, Andrea
author_facet Celentano, Michael
Cheng, Chen
Montanari, Andrea
contents We study a class of deterministic flows in ${\mathbb R}^{d\times k}$, parametrized by a random matrix ${\boldsymbol X}\in {\mathbb R}^{n\times d}$ with i.i.d. centered subgaussian entries. We characterize the asymptotic behavior of these flows over bounded time horizons, in the high-dimensional limit in which $n,d\to\infty$ with $k$ fixed and converging aspect ratios $n/d\toδ$. The asymptotic characterization we prove is in terms of a system of nonlinear stochastic processes in $k$ dimensions, whose parameters are determined by a fixed point condition. This type of characterization is known in physics as dynamical mean field theory. Rigorous results of this type have been obtained in the past for a few spin glass models. Our proof is based on time discretization and a reduction to certain iterative schemes known as approximate message passing (AMP) algorithms, as opposed to earlier work that was based on large deviations theory and stochastic processes theory. The new approach provides a unified view of a general class of algorithms and implies that the high-dimensional behavior of the flow is universal with respect to the distribution of the entries of ${\boldsymbol X}$. As specific applications, we obtain high-dimensional characterizations of gradient flow in some classical models from statistics and machine learning, under a random design assumption.
format Preprint
id arxiv_https___arxiv_org_abs_2112_07572
institution arXiv
publishDate 2021
record_format arxiv
spellingShingle The high-dimensional asymptotics of first order methods with random data
Celentano, Michael
Cheng, Chen
Montanari, Andrea
Probability
Statistics Theory
We study a class of deterministic flows in ${\mathbb R}^{d\times k}$, parametrized by a random matrix ${\boldsymbol X}\in {\mathbb R}^{n\times d}$ with i.i.d. centered subgaussian entries. We characterize the asymptotic behavior of these flows over bounded time horizons, in the high-dimensional limit in which $n,d\to\infty$ with $k$ fixed and converging aspect ratios $n/d\toδ$. The asymptotic characterization we prove is in terms of a system of nonlinear stochastic processes in $k$ dimensions, whose parameters are determined by a fixed point condition. This type of characterization is known in physics as dynamical mean field theory. Rigorous results of this type have been obtained in the past for a few spin glass models. Our proof is based on time discretization and a reduction to certain iterative schemes known as approximate message passing (AMP) algorithms, as opposed to earlier work that was based on large deviations theory and stochastic processes theory. The new approach provides a unified view of a general class of algorithms and implies that the high-dimensional behavior of the flow is universal with respect to the distribution of the entries of ${\boldsymbol X}$. As specific applications, we obtain high-dimensional characterizations of gradient flow in some classical models from statistics and machine learning, under a random design assumption.
title The high-dimensional asymptotics of first order methods with random data
topic Probability
Statistics Theory
url https://arxiv.org/abs/2112.07572