Research map / Adversarial robustness and security / Certified and empirical robustness
Certified Defenses for 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 the candidate attack generated by the online learning bound rather than label flipping | Oracle sphere and slab defense on MNIST-1-7 and Dogfish · MNIST-1-7 and Dogfish | test loss | Helped |
| Replace the candidate attack with gradient descent on test loss | Oracle sphere and slab defense · MNIST-1-7 and Dogfish | test loss | Hurt |
| Replace the candidate attack with label flipping | Oracle sphere and slab defense · MNIST-1-7 and Dogfish | test loss | Hurt |
| Add 30% poisoned data under the oracle sphere and slab defense | Binary SVM on MNIST-1-7 · MNIST-1-7 | test error: at most 4% | No clear effect |
| Add 30% poisoned data under the oracle sphere and slab defense | Binary SVM on Dogfish · Dogfish | test error: at most 4% | No clear effect |
| Add 3% poisoned data under the oracle sphere and slab defense | Binary SVM on IMDB · IMDB | test error: 12% to 23% | Hurt |
| Use empirical poisoned class centroids instead of true class centroids (ablation) | Sphere and slab defense on MNIST-1-7 · MNIST-1-7 | test 0-1 loss: 0.01 to 0.40 | Hurt |
| Use empirical poisoned class centroids instead of true class centroids (ablation) | Sphere and slab defense on Dogfish · Dogfish | test 0-1 loss: 0.01 to 0.22 | Hurt |
| Increase poisoning to 30% rather than keeping it at 5% or less (ablation) | Data-dependent sphere and slab defense on MNIST-1-7 and Dogfish · MNIST-1-7 and Dogfish | train loss | Hurt |
| Use the data-dependent defense instead of the oracle defense (ablation) | Sphere and slab defense on MNIST-1-7 · MNIST-1-7 | test 0-1 loss: 0.01 to 0.40 | Hurt |
| Use the data-dependent defense instead of the oracle defense (ablation) | Sphere and slab defense on Dogfish · Dogfish | test 0-1 loss: 0.01 to 0.22 | Hurt |
| Add 30% poisoned data under the data-dependent sphere and slab defense | Binary SVM on MNIST-1-7 · MNIST-1-7 | hinge loss: 0.03 to 0.69 | Hurt |
Closest papers by meaning
- On Robustness of Linear Classifiers to Targeted Data Poisoning (AAAI 2026 · similarity 0.90)
- Not All Poisons are Created Equal: Robust Training against Data Poisoning (ICML 2022 · similarity 0.89)
- Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samples (NeurIPS 2026 · similarity 0.88)
- Attack To Defend: Exploiting Adversarial Attacks for Detecting Poisoned Models (CVPR 2024 · similarity 0.87)
- Phantom Transfer: Data Poisoning can Survive Data-Level Defences (NeurIPS 2026 · similarity 0.87)
- Sharpness-Aware Data Poisoning Attack (ICLR 2024 · 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.