Research map / Adversarial robustness and security / Certified and empirical robustness
Intrinsic Certified Robustness of Bagging against Data Poisoning Attacks
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 |
|---|---|---|---|
| Use bagging with subsamples of k=100 and N=1,000 base classifiers | BaggingCertify · MNIST | certified accuracy at 100 poisoned training examples: 91.1% | Helped |
| Use bagging certification instead of existing certified defenses | MNIST 1/7 · MNIST 1/7, 50 poisoned training examples | certified accuracy: 96.95% vs 0 | Helped |
| Use transfer learning with an ImageNet-pretrained Inception-v3 feature extractor (ablation) | BaggingCertify base classifiers · CIFAR10 | certified accuracy | Helped |
| Increase subsample size k (ablation) | BaggingCertify · MNIST and CIFAR10 | certified accuracy | Mixed |
| Increase confidence parameter alpha (ablation) | BaggingCertify · MNIST and CIFAR10 | certified accuracy | Helped |
| Increase number of sampled base classifiers N (ablation) | BaggingCertify · MNIST and CIFAR10 | certified accuracy | Helped |
| Restrict the attacker to deleting training examples (ablation) | BaggingCertify · MNIST and CIFAR10 | certified accuracy | Helped |
| Allow modification attacks, or allow all attack types instead of deletion only (ablation) | BaggingCertify · MNIST and CIFAR10 | certified accuracy | Hurt |
Closest papers by meaning
- On Collective Robustness of Bagging Against Data Poisoning (ICML 2022 · similarity 0.93)
- BagFlip: A Certified Defense Against Data Poisoning (NeurIPS 2022 · similarity 0.87)
- Robust Conformal Outlier Detection under Contaminated Reference Data (ICML 2025 · similarity 0.83)
- Provably Reliable Conformal Prediction Sets in the Presence of Data Poisoning (ICLR 2025 · similarity 0.82)
- Naive Bayes Classifiers over Missing Data: Decision and Poisoning (ICML 2024 · similarity 0.82)
- Like Oil and Water: Group Robustness Methods and Poisoning Defenses May Be at Odds (ICLR 2024 · similarity 0.82)
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.