Research map / Adversarial robustness and security / Black-box and transfer attacks
On the Perils of Cascading Robust Classifiers
Ravi Mangal, Zifan Wang, Chi Zhang, Klas Leino, Corina Pasareanu, Matt Fredrikson
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Apply CasA to the cascading ensemble | MNIST ℓ∞ ε=0.1 Small Exact cascade · MNIST ℓ∞ ε=0.1 | FPR (%): 88.71 | Hurt |
| Apply CasA to the cascading ensemble | MNIST ℓ∞ ε=0.1 Small Exact cascade · MNIST ℓ∞ ε=0.1 | ERA (%): 11.17 | Hurt |
| Use cascading instead of the single best constituent | CIFAR-10 ℓ∞ ε=2/255 Large · CIFAR-10 ℓ∞ ε=2/255 | ERA (%): 52.65 -> 58.15 | Hurt |
| Use sound weighted voting instead of the single model | MNIST ℓ∞ ε=0.1 Small · MNIST ℓ∞ ε=0.1 | CRA (%): 94.94 -> 94.94 | No clear effect |
| Use uniform voting instead of the single model | MNIST ℓ∞ ε=0.1 Small · MNIST ℓ∞ ε=0.1 | CRA (%): 94.94 -> 9.29 | Hurt |
| Use uniform voting instead of the single model | CIFAR-10 ℓ∞ ε=2/255 Small · CIFAR-10 ℓ∞ ε=2/255 | CRA (%): 46.43 -> 18.58 | Hurt |
| Replace dual-network attack objectives with surrogate objectives (ablation) | CasA on MNIST ℓ∞ ε=0.3 Small cascade · MNIST ℓ∞ ε=0.3 | ERA (%): 4.93 -> 7.67 | Mixed |
| Replace dual-network attack objectives with surrogate objectives (ablation) | CasA on CIFAR-10 ℓ∞ ε=2/255 Small cascade · CIFAR-10 ℓ∞ ε=2/255 | ERA (%): 23.79 -> 22.92 | Mixed |
| Train constituents non-sequentially on the full training dataset (ablation) | Uniform voting ensemble · CIFAR-10 ℓ∞ ε=2/255 Small | CRA (%): 18.58 -> 43.48 | Mixed |
| Train constituents non-sequentially on the full training dataset (ablation) | Uniform voting ensemble · CIFAR-10 ℓ∞ ε=2/255 Small | CRA (%): 46.43 -> 43.48 | Hurt |
| Use weighted voting with learned weights | Non-sequentially trained ensemble · CIFAR-10 ℓ∞ ε=2/255 Small | CRA (%): 46.43 -> 42.35 | Hurt |
Closest papers by meaning
- On the Certified Robustness for Ensemble Models and Beyond (ICLR 2022 · similarity 0.91)
- Certifying Ensembles: A General Certification Theory with S-Lipschitzness (ICML 2023 · similarity 0.88)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.87)
- Training Robust Ensembles Requires Rethinking Lipschitz Continuity (ICLR 2025 · similarity 0.87)
- Tight Certificates of Adversarial Robustness for Randomly Smoothed Classifiers (NeurIPS 2019 · similarity 0.86)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.86)
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.
