Theoretical Lower Bounds on the Robustness of Deep ReLU Networks

arXiv:2602.18674 · cs.LG, cs.NE · Submitted 2026-02-21 · Read on arXiv

cs.LG, cs.NE

Submitted: 2026-02-21

Updated: 2026-10-02

Terminology

• Robustness — 4× in this paper · explained in 5 episodes

Sources

• Intriguing properties of neural networks
• Attacks in Adversarial Machine Learning: A Systematic Survey from the Life-cycle Perspective
• Feature compression is the root cause of adversarial fragility in neural network classifiers
• Explaining and Harnessing Adversarial Examples
• Talagrand's mathematical journey to the Abel Prize 2024

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks