Research map / Learning theory and algorithms / Prediction and decision losses
(De-)Randomized Smoothing for Decision Stump Ensembles
Miklós Z. Horváth, Mark Niklas Mueller, Marc Fischer, Martin Vechev
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 |
|---|---|---|---|
| DRS vs prior state-of-the-art, independent stumps | tree ensembles · MNIST 2 vs. 6 (l2, r=0.8) | certified accuracy (%): 23.0 -> 89.6 | Improved |
| DRS vs prior state-of-the-art natural accuracy | tree ensembles · MNIST 2 vs. 6 | natural accuracy (%): 86.3 -> 96.3 | Improved |
| add boosting to independently trained stumps (ablation) | DRS · MNIST 1 vs. 5 (l1, r=1.0) | certified accuracy (%): 94.1 -> 98.1 | Improved |
| add boosting to independently trained stumps (ablation) | DRS · BreastCancer (l2, r=0.7) | certified accuracy (%): 75.2 -> 82.5 | Improved |
| add boosting to independently trained stumps (ablation) | DRS · MNIST 2 vs. 6 (l2, r=0.8) | certified accuracy (%): 89.6 -> 89.6 | No clear effect |
| add boosting to independently trained stumps (ablation) | DRS · Diabetes (l1, r=0.05) | certified accuracy (%): 69.5 -> 72.1 | Improved |
| use categorical features in addition to numerical (ablation) | DRS · Adult (l2 r=0.75) | balanced certified accuracy (%): 27.4 -> 63.3 | Improved |
| use categorical features in addition to numerical (ablation) | DRS · Credit (l2 r=1.00) | balanced certified accuracy (%): 9.7 -> 57.1 | Improved |
| adversarially perturb 3 categorical features (l0 radius 3 vs 0) (ablation) | DRS · Adult (l2 r=0.75) | balanced certified accuracy (%): 63.3 -> 15.4 | Hurt |
| deterministic DRS instead of sampling-based RS (n=100 000) (ablation) | DRS · MNIST 1 vs. 5 | ACR: 33.1% | Improved |
| train with MLE criterion instead of sampling perturbed data (ablation) | DRS · BreastCancer sigma=1 | ACR: 0.567 -> 0.675 | Improved |
| train with MLE criterion instead of clean data (ablation) | DRS · BreastCancer sigma=1 | ACR: 0.356 -> 0.675 | Improved |
Closest papers by meaning
- Boosting Randomized Smoothing with Variance Reduced Classifiers (ICLR 2022 · similarity 0.88)
- On Lp-norm Robustness of Ensemble Decision Stumps and Trees (ICML 2020 · similarity 0.88)
- Hierarchical Randomized Smoothing (NeurIPS 2023 · similarity 0.88)
- Multistage Defer Trees for Hybrid Interpretability: If at First You Can't Succeed, Tree Again (NeurIPS 2026 · similarity 0.87)
- Generalized additive models via direct optimization of regularized decision stump forests (ICML 2025 · similarity 0.86)
- Clustered Randomized Smoothing for Stochastic Prediction Functions (NeurIPS 2026 · similarity 0.85)
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.
