Research map / Adversarial robustness and security / Certified and empirical robustness
Verifying Robustness of Gradient Boosted Models
Gil Einziger, Maayan Goldstein, Yaniv Sa’ar, Itai Segall
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 |
|---|---|---|---|
| Use smaller tree depth while keeping model accuracy similar | Gradient boosted models · MNIST, ε=1 | verified robust fraction (ρ): 53% -> 27.5% | Improved |
| Increase tree depth from 4 to 20 in similarly accurate models | Gradient boosted models · MNIST, ε=1 | verified robust fraction (ρ): 53% -> 27.5% | Hurt |
| Increase tree count from 50 to 100 at depth 8 | Gradient boosted model · MNIST, ε=1 | verified robust fraction (ρ): 53.5% -> 29% | Mixed |
| Increase tree depth from 3 to 8 at 50 trees | Gradient boosted model · MNIST, ε=1 | verified robust fraction (ρ): 24% -> 53.5% | Improved |
| Increase tree count from 20 to 100 at depth 3 | Gradient boosted model · MNIST, ε=1 | verified robust fraction (ρ): 16.5% -> 39.5% | Improved |
| Increase tree count from 20 to 100 at depth 10 | Gradient boosted model · MNIST, ε=1 | verified robust fraction (ρ): 39.5% -> 15% | Hurt |
| Increase tree depth | Gradient boosted regressors on HSKC · HSKC | robust fraction (ρ) | Hurt |
| Increase number of trees | Gradient boosted regressors on HSKC · HSKC | robust fraction (ρ) | Hurt |
| Increase perturbation bound ε | Gradient boosted classifiers · MNIST | verified robust fraction (ρ): 39.5% -> 10% | Hurt |
Closest papers by meaning
- Versatile Verification of Tree Ensembles (ICML 2021 · similarity 0.87)
- LEVIS: Large Exact Verifiable Input Spaces for Neural Networks (ICML 2025 · similarity 0.86)
- Stress-Testing Neural Network Verifiers with Provably Robust Instances (NeurIPS 2026 · similarity 0.85)
- Certified but Private: Scalable Zero-Knowledge Proofs for the Formal Verification of Neural Networks (NeurIPS 2026 · similarity 0.85)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.85)
- Sparse Transfer Learning Accelerates and Enhances Certified Robustness: A Comprehensive Study (AAAI 2025 · similarity 0.84)
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.
