Research

Paper

AI LLM February 19, 2026

Convergence Analysis of Two-Layer Neural Networks under Gaussian Input Masking

Authors

Afroditi Kolomvaki, Fangshuo Liao, Evan Dramko, Ziyun Guang, Anastasios Kyrillidis

Abstract

We investigate the convergence guarantee of two-layer neural network training with Gaussian randomly masked inputs. This scenario corresponds to Gaussian dropout at the input level, or noisy input training common in sensor networks, privacy-preserving training, and federated learning, where each user may have access to partial or corrupted features. Using a Neural Tangent Kernel (NTK) analysis, we demonstrate that training a two-layer ReLU network with Gaussian randomly masked inputs achieves linear convergence up to an error region proportional to the mask's variance. A key technical contribution is resolving the randomness within the non-linear activation, a problem of independent interest.

Metadata

arXiv ID: 2602.17423
Provider: ARXIV
Primary Category: cs.LG
Published: 2026-02-19
Fetched: 2026-02-21 18:51

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.17423v1</id>\n    <title>Convergence Analysis of Two-Layer Neural Networks under Gaussian Input Masking</title>\n    <updated>2026-02-19T14:55:10Z</updated>\n    <link href='https://arxiv.org/abs/2602.17423v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.17423v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>We investigate the convergence guarantee of two-layer neural network training with Gaussian randomly masked inputs. This scenario corresponds to Gaussian dropout at the input level, or noisy input training common in sensor networks, privacy-preserving training, and federated learning, where each user may have access to partial or corrupted features. Using a Neural Tangent Kernel (NTK) analysis, we demonstrate that training a two-layer ReLU network with Gaussian randomly masked inputs achieves linear convergence up to an error region proportional to the mask's variance. A key technical contribution is resolving the randomness within the non-linear activation, a problem of independent interest.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.LG'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.DS'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='math.OC'/>\n    <published>2026-02-19T14:55:10Z</published>\n    <arxiv:comment>69 pages, submitted to AI/ML Journal</arxiv:comment>\n    <arxiv:primary_category term='cs.LG'/>\n    <author>\n      <name>Afroditi Kolomvaki</name>\n    </author>\n    <author>\n      <name>Fangshuo Liao</name>\n    </author>\n    <author>\n      <name>Evan Dramko</name>\n    </author>\n    <author>\n      <name>Ziyun Guang</name>\n    </author>\n    <author>\n      <name>Anastasios Kyrillidis</name>\n    </author>\n  </entry>"
}