Saved in:
Bibliographic Details
Main Authors: Zhang, Chen, Cheng, Yuxin, Ding, Chenchen, Wang, Shuqi, Lei, Jingreng, Yu, Runsheng, WU, Yik-Chung, Wong, Ngai
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.21169
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917356617334784
author Zhang, Chen
Cheng, Yuxin
Ding, Chenchen
Wang, Shuqi
Lei, Jingreng
Yu, Runsheng
WU, Yik-Chung
Wong, Ngai
author_facet Zhang, Chen
Cheng, Yuxin
Ding, Chenchen
Wang, Shuqi
Lei, Jingreng
Yu, Runsheng
WU, Yik-Chung
Wong, Ngai
contents Zeroth-order (ZO) optimization enables memory-efficient training of neural networks by estimating gradients via forward passes only, eliminating the need for backpropagation. However, the stochastic nature of gradient estimation significantly obscures the training dynamics, in contrast to the well-characterized behavior of first-order methods under Neural Tangent Kernel (NTK) theory. To address this, we introduce the Neural Zeroth-order Kernel (NZK) to describe model evolution in function space under ZO updates. For linear models, we prove that the expected NZK remains constant throughout training and depends explicitly on the first and second moments of the random perturbation directions. This invariance yields a closed-form expression for model evolution under squared loss. We further extend the analysis to linearized neural networks. Interpreting ZO updates as kernel gradient descent via NZK provides a novel perspective for potentially accelerating convergence. Extensive experiments across synthetic and real-world datasets (including MNIST, CIFAR-10, and Tiny ImageNet) validate our theoretical results and demonstrate acceleration when using a single shared random vector.
format Preprint
id arxiv_https___arxiv_org_abs_2603_21169
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Model Evolution Under Zeroth-Order Optimization: A Neural Tangent Kernel Perspective
Zhang, Chen
Cheng, Yuxin
Ding, Chenchen
Wang, Shuqi
Lei, Jingreng
Yu, Runsheng
WU, Yik-Chung
Wong, Ngai
Machine Learning
Zeroth-order (ZO) optimization enables memory-efficient training of neural networks by estimating gradients via forward passes only, eliminating the need for backpropagation. However, the stochastic nature of gradient estimation significantly obscures the training dynamics, in contrast to the well-characterized behavior of first-order methods under Neural Tangent Kernel (NTK) theory. To address this, we introduce the Neural Zeroth-order Kernel (NZK) to describe model evolution in function space under ZO updates. For linear models, we prove that the expected NZK remains constant throughout training and depends explicitly on the first and second moments of the random perturbation directions. This invariance yields a closed-form expression for model evolution under squared loss. We further extend the analysis to linearized neural networks. Interpreting ZO updates as kernel gradient descent via NZK provides a novel perspective for potentially accelerating convergence. Extensive experiments across synthetic and real-world datasets (including MNIST, CIFAR-10, and Tiny ImageNet) validate our theoretical results and demonstrate acceleration when using a single shared random vector.
title Model Evolution Under Zeroth-Order Optimization: A Neural Tangent Kernel Perspective
topic Machine Learning
url https://arxiv.org/abs/2603.21169