![[2602.17948] A Geometric Probe of the Accuracy-Robustness Trade-off: Sharp Boundaries in Symmetry-Breaking Dimensional Expansion](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.17948] A Geometric Probe of the Accuracy-Robustness Trade-off: Sharp Boundaries in Symmetry-Breaking Dimensional Expansion
Summary
This paper explores the accuracy-robustness trade-off in deep learning through a geometric lens, utilizing Symmetry-Breaking Dimensional Expansion to enhance accuracy while revealing vulnerabilities to adversarial attacks.
Why It Matters
Understanding the accuracy-robustness trade-off is crucial for developing more resilient machine learning models. This research provides insights into how geometric alterations can improve performance while highlighting the inherent risks, informing future model design and training strategies.
Key Takeaways
- Symmetry-Breaking Dimensional Expansion can enhance model accuracy significantly.
- Increased accuracy may lead to reduced robustness against adversarial attacks.
- The study reveals that sharp boundaries in the optimization landscape contribute to model fragility.
- Test-time mask projection can mitigate vulnerabilities introduced by dimensional expansion.
- The findings offer a geometric explanation for the accuracy-robustness paradox in deep learning.
Computer Science > Machine Learning arXiv:2602.17948 (cs) [Submitted on 20 Feb 2026] Title:A Geometric Probe of the Accuracy-Robustness Trade-off: Sharp Boundaries in Symmetry-Breaking Dimensional Expansion Authors:Yu Bai, Zhe Wang, Jiarui Zhang, Dong-Xiao Zhang, Yinjun Gao, Jun-Jie Zhang View a PDF of the paper titled A Geometric Probe of the Accuracy-Robustness Trade-off: Sharp Boundaries in Symmetry-Breaking Dimensional Expansion, by Yu Bai and 5 other authors View PDF HTML (experimental) Abstract:The trade-off between clean accuracy and adversarial robustness is a pervasive phenomenon in deep learning, yet its geometric origin remains elusive. In this work, we utilize Symmetry-Breaking Dimensional Expansion (SBDE) as a controlled probe to investigate the mechanism underlying this trade-off. SBDE expands input images by inserting constant-valued pixels, which breaks translational symmetry and consistently improves clean accuracy (e.g., from $90.47\%$ to $95.63\%$ on CIFAR-10 with ResNet-18) by reducing parameter degeneracy. However, this accuracy gain comes at the cost of reduced robustness against iterative white-box attacks. By employing a test-time \emph{mask projection} that resets the inserted auxiliary pixels to their training values, we demonstrate that the vulnerability stems almost entirely from the inserted dimensions. The projection effectively neutralizes the attacks and restores robustness, revealing that the model achieves high accuracy by creating \emph{s...