Research map / Adversarial robustness and security / Attacks on text and graphs
Friendly Noise against Adversarial Noise: A Powerful Defense against Data Poisoning Attack
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 |
|---|---|---|---|
| Add friendly and varying Bernoulli noise | ResNet-18 trained on CIFAR-10 · Gradient Matching (ξ=16) | Poison Acc: 75.00 -> 0.00 | Helped |
| Add friendly and varying Bernoulli noise | ResNet-18 trained on CIFAR-10 · Gradient Matching (ξ=16) | Test Acc: 93.50 -> 91.52 | Hurt |
| Combine friendly noise with varying random noise rather than using friendly noise alone (ablation) | FrieNDs on ResNet-18 · Gradient Matching (ξ=16) | Poison Acc: 10.00 -> 0.00 | Helped |
| Combine friendly noise with smaller Gaussian noise rather than using Gaussian noise alone (ablation) | FrieNDs on ResNet-18 · Gradient Matching (ξ=16) | Test Acc: 89.46 -> 91.50 | Helped |
| Use friendly noise without varying random noise (ablation) | FrieNDs on ResNet-18 · Gradient Matching (ξ=16) | Poison Acc: 0.00 -> 10.00 | Hurt |
| Replace Bernoulli random noise with uniform noise (ablation) | FrieNDs on ResNet-18 · Gradient Matching (ξ=16) | Poison Acc: 0.00 -> 5.00 | Hurt |
| Replace the L2 friendly-noise objective with L1 (ablation) | FrieNDs-B on ResNet-18 · Gradient Matching (ξ=16) | Test Acc: 91.52% -> 91.50% | No clear effect |
| Replace the L2 friendly-noise objective with L∞ (ablation) | FrieNDs-B on ResNet-18 · Gradient Matching (ξ=16) | Test Acc: 91.52% -> 91.37% | No clear effect |
| Use FrieNDs-B against Sleeper Agent | Sleeper Agent defense on 24 datasets · Sleeper Agent | Poison Acc: 83.48% -> 21.52% | Helped |
| Use FrieNDs-B against Sleeper Agent | Sleeper Agent defense on 24 datasets · Sleeper Agent | Test Acc: 91.56% -> 89.76% | Hurt |
| Adapt Sleeper Agent poison generation to Bernoulli noise | FrieNDs-B · Adaptive Sleeper Agent attack | Poison Acc: 21.52 ± 8.39 -> 31.05 ± 9.45 | Mixed |
| Adapt Sleeper Agent poison generation to FrieNDs-B | FrieNDs-B · Adaptive Sleeper Agent attack | Poison Acc: 21.52 ± 8.39 -> 30.50 ± 9.40 | Mixed |
Closest papers by meaning
- Re-thinking Data Availability Attacks Against Deep Neural Networks (CVPR 2024 · similarity 0.89)
- Attack To Defend: Exploiting Adversarial Attacks for Detecting Poisoned Models (CVPR 2024 · similarity 0.88)
- Generalization Bound and New Algorithm for Clean-Label Backdoor Attack (ICML 2024 · similarity 0.88)
- Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks (ICLR 2025 · similarity 0.88)
- Mind the Gap: Detecting Black-box Adversarial Attacks in the Making through Query Update Analysis (CVPR 2025 · similarity 0.87)
- On the Adversarial Risk of Test Time Adaptation: An Investigation into Realistic Test-Time Data Poisoning (ICLR 2025 · similarity 0.87)
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.