Research map / Adversarial robustness and security / Certified and empirical robustness
Provably robust boosted decision stumps and trees against adversarial attacks
Maksym Andriushchenko, Matthias Hein
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 |
|---|---|---|---|
| Train boosted stumps by minimizing the exact robust loss rather than using standard training | Boosted decision stumps · breast-cancer (ε∞=0.3) | RTE (%): 98.5 -> 10.9 | Improved |
| Train boosted stumps with the robust-loss bound rather than standard training | Boosted decision stumps · MNIST 2-6 (ε∞=0.3) | RTE (%): 99.9 -> 9.1 | Improved |
| Use robust-loss-bound stump training instead of Robust stumps [cite] | Boosted decision stumps · diabetes (ε∞=0.05) | RTE (%): 30.5 -> 33.1 | Hurt |
| Use robust-loss-bound stump training instead of Robust stumps [cite] | Boosted decision stumps · Results across eight binary datasets | RTE (%) | Mixed |
| Train boosted trees by minimizing the upper bound on robust loss rather than using standard training | Boosted decision trees · breast-cancer (ε∞=0.3) | RTE (%): 81.0 -> 6.6 | Improved |
| Use robust-loss-bound tree training instead of Robust trees [cite] | Boosted decision trees · Results across eight binary datasets | RTE (%) | Improved |
| Use robust-loss-bound tree training rather than adversarial training | Boosted decision trees · GTS 30-70 (ε∞=8/255) | RTE (%): 26.7 -> 20.9 | Improved |
| Increase weak-learner expressiveness from stumps to depth-4 trees | Robust boosted models · breast-cancer (ε∞=0.3) | RTE (%): 10.9 -> 6.6 | Improved |
| Use the tree-wise upper bound on robust test error rather than exact MIP evaluation | Our robust boosted trees · Results across binary datasets | RTE (%) | No clear effect |
| Train stumps by optimizing the exact robust loss rather than its stump-wise bound (ablation) | Our robust boosted stumps · RTE across eight binary datasets | RTE (%) | No clear effect |
| Use robust-loss-bound stump training instead of exact robust-loss stump training (ablation) | Our robust boosted stumps · MNIST 1-5 (ε∞=0.3) | RTE (%): 3.6 -> 3.7 | Hurt |
| Increase perturbation radius ε during robust training (ablation) | Robust boosted trees · Robustness–accuracy tradeoff | test error | Mixed |
Closest papers by meaning
- Fast Provably Robust Decision Trees and Boosting (ICML 2022 · similarity 0.92)
- Robust Decision Trees Against Adversarial Examples (ICML 2019 · similarity 0.89)
- TTTS: Tree Test Time Simulation for Enhancing Decision Tree Robustness against Adversarial Examples (AAAI 2024 · similarity 0.87)
- Robust Loss Functions for Training Decision Trees with Noisy Labels (AAAI 2024 · similarity 0.86)
- Accelerating Adversarially Robust Model Selection for Deep Neural Networks via Racing (AAAI 2024 · similarity 0.86)
- Expressive Losses for Verified Robustness via Convex Combinations (ICLR 2024 · 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.
