Wasserstein-Robust Training for One-Hidden-Layer ReLU Networks with Distributional Guarantees
Electronics · 2026-05-27 · 15 (11) · 2316
DOI: 10.3390/electronics15112316 ↗At a glance
Journal article in Electronics, published online on 2026-05-27. Written by Yang Wang and 2 co-authors. 54 references. The full text is freely available (open access, CC BY 4.0). Funded by: Department of Education of Guangdong Province.
Abstract (original language)
Neural networks are vulnerable to adversarial perturbations, which motivates training procedures with formal robustness guarantees. In this paper, we study one-hidden-layer ReLU networks from the perspective of Wasserstein distributional robustness. Leveraging the network structure, we derive an upper bound for the intractable robust surrogate in the form of a tractable regularized empirical risk objective whose regularizer is computed through a low-rank optimization problem based on Burer–Monteiro factorization. This reformulation yields a distributional robustness certificate on the worst-case expected loss over a Wasserstein ball. The upper bound construction and distributional certificate are developed for the shallow fixed-output multiclass formulation, while the optimization analysis focuses on a binary specialization with margin loss and exact linear separability. We also analyze a modified stochastic gradient descent scheme for the resulting regularized problem in this binary linearly separable setting, and we establish a corresponding generalization bound. The experiments validate the proposed surrogate and training procedure on binary MNIST and CIFAR-10 tasks, and we added a 10-class MNIST experiment to further check the multiclass trainability of the surrogate.
Funding
- Department of Education of Guangdong Province
Cite
APA
Wang, Y., Zou, D., & Liu, X. (2026). Wasserstein-Robust Training for One-Hidden-Layer ReLU Networks with Distributional Guarantees. Electronics, 15(11), 2316. https://doi.org/10.3390/electronics15112316
Vancouver
Wang Y, Zou D, Liu X. Wasserstein-Robust Training for One-Hidden-Layer ReLU Networks with Distributional Guarantees. Electronics. 2026;15(11):2316. doi:10.3390/electronics15112316
BibTeX
@article{Wang2026,
author = {Wang, Yang and Zou, Difan and Liu, Xiaona},
title = {Wasserstein-Robust Training for One-Hidden-Layer ReLU Networks with Distributional Guarantees},
journal = {Electronics},
year = {2026},
volume = {15},
number = {11},
pages = {2316},
doi = {10.3390/electronics15112316}
}
EUVIMED