Research map / Adversarial robustness and security / Certified and empirical robustness
On the Generalization Analysis of Adversarial Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace additive adversarial noise with general (non-additive) noise application functions in the adversarial generalization analysis | Adversarial learning generalization bounds · Theoretical analysis (multi-class linear and neural networks) | adversarial generalization bound scaling | Helped |
| Scale of bound in number of classes K changed from O(sqrt(K)) or O(K) in prior work to O(log(K)) via covering-number analysis | Multi-class linear classifiers under l-infinity additive attack · Theoretical analysis (K classes) | dependence on number of classes | Helped |
| Optimistic (local Rademacher) bounds for smooth robust-self-bounding losses, giving fast rates when empirical adversarial risk is zero | Adversarial deep learning with smooth margin loss · Theoretical analysis | generalization rate | Helped |
| Bound dependence on input dimension d for additive l-infinity attacks includes extra sqrt(d) factor from noise-set complexity, unlike prior bounds | Multi-class linear classifiers under l-infinity additive attack · Theoretical analysis | dependence on input dimension | Hurt |
Closest papers by meaning
- Algorithmic Stability Based Generalization Bounds for Adversarial Training (ICLR 2025 · similarity 0.90)
- Stability Analysis and Generalization Bounds of Adversarial Training (NeurIPS 2022 · similarity 0.90)
- Theoretical Analysis of Adversarial Learning: A Minimax Approach (NeurIPS 2019 · similarity 0.90)
- More Data Can Expand The Generalization Gap Between Adversarially Robust and Standard Models (ICML 2020 · similarity 0.90)
- Generalizability of Neural Networks Minimizing Empirical Risk Based on Expressive Power (ICLR 2025 · similarity 0.88)
- Stability and Generalization of Adversarial Training for Shallow Neural Networks with Smooth Activation (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.