Weakly Supervised Contrastive Adversarial Training for Learning Robust Features from Semi-supervised Data

Lili Zhang, Chengpei Wu, Ning Yang
2/10/2026

No official code found. Best implementation: zhang-lilin/WSCAT (0 stars, last commit 2026-01-14).

Abstract

Existing adversarial training (AT) methods often suffer from incomplete perturbation, meaning that not all non-robust features are perturbed when generating adversarial examples (AEs). This results in residual correlations between non-robust features and labels, leading to suboptimal learning of robust features. However, achieving complete perturbation—perturbing as many non-robust features as possible, is challenging due to the difficulty in distinguishing robust and non-robust features and the sparsity of labeled data. To address these challenges, we propose a novel approach called Weakly Supervised Contrastive Adversarial Training (WSCAT). WSCAT ensures complete perturbation for improved learning of robust features by disrupting correlations between non-robust features and labels through complete AE generation over partially labeled data, grounded in information theory. Extensive theoretical analysis and comprehensive experiments on widely adopted benchmarks validate the superiority of WSCAT. Our code is available at https://github.com/zhang-lilin/WSCAT.

View on arXivView PDF

Code Implementations(1)

zhang-lilin/WSCATCommunity71%

Implementation code for "Weakly Supervised Contrastive Adversarial Training for Learning Robust Features from Semi-supervised Data".

01PythonMar 20, 20259 months ago

Similar Papers

Self-Supervised Adversarial Training via Diverse Augmented Queries and Self-Supervised Double Perturbation

88% match1 repo2026

On the Duality Between Sharpness-Aware Minimization and Adversarial Training

87% match1 repo252026

Enhancing Adversarial Robustness in Low-Label Regime via Adaptively Weighted Regularization and Knowledge Distillation

86% match1 repo12026

Salient Information Preserving Adversarial Training Improves Clean and Robust Accuracy

85% match2025

Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data

85% match2026

Effective and Robust Adversarial Training against Data and Label Corruptions

84% match2024

Cite this paper

@article{zhang2026weakly,
  title  = {Weakly Supervised Contrastive Adversarial Training for Learning Robust Features from Semi-supervised Data},
  author = {Lili Zhang and Chengpei Wu and Ning Yang},
  year   = {2026},
  eprint = {2503.11032},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/2503.11032},
  journal = {CVPR 2025 2025}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2503.11032.svg)](https://codeofpaper.com/paper/2503.11032)

Discussion