Saved in:
Bibliographic Details
Main Authors: Li, Bingcong, Zhang, Liang, Mokhtari, Aryan, He, Niao
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.18965
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910743472898048
author Li, Bingcong
Zhang, Liang
Mokhtari, Aryan
He, Niao
author_facet Li, Bingcong
Zhang, Liang
Mokhtari, Aryan
He, Niao
contents This work revisits the classical low-rank matrix factorization problem and unveils the critical role of initialization in shaping convergence rates for such nonconvex and nonsmooth optimization. We introduce Nystrom initialization, which significantly improves the global convergence of Scaled Gradient Descent (ScaledGD) in both symmetric and asymmetric matrix factorization tasks. Specifically, we prove that ScaledGD with Nystrom initialization achieves quadratic convergence in cases where only linear rates were previously known. Furthermore, we extend this initialization to low-rank adapters (LoRA) commonly used for finetuning foundation models. Our approach, NoRA, i.e., LoRA with Nystrom initialization, demonstrates superior performance across various downstream tasks and model scales, from 1B to 7B parameters, in large language and diffusion models.
format Preprint
id arxiv_https___arxiv_org_abs_2410_18965
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On the Crucial Role of Initialization for Matrix Factorization
Li, Bingcong
Zhang, Liang
Mokhtari, Aryan
He, Niao
Machine Learning
Signal Processing
Optimization and Control
This work revisits the classical low-rank matrix factorization problem and unveils the critical role of initialization in shaping convergence rates for such nonconvex and nonsmooth optimization. We introduce Nystrom initialization, which significantly improves the global convergence of Scaled Gradient Descent (ScaledGD) in both symmetric and asymmetric matrix factorization tasks. Specifically, we prove that ScaledGD with Nystrom initialization achieves quadratic convergence in cases where only linear rates were previously known. Furthermore, we extend this initialization to low-rank adapters (LoRA) commonly used for finetuning foundation models. Our approach, NoRA, i.e., LoRA with Nystrom initialization, demonstrates superior performance across various downstream tasks and model scales, from 1B to 7B parameters, in large language and diffusion models.
title On the Crucial Role of Initialization for Matrix Factorization
topic Machine Learning
Signal Processing
Optimization and Control
url https://arxiv.org/abs/2410.18965