Research map / Adversarial robustness and security / Attacks on text and graphs
Accumulative Poisoning Attacks on Real-time Data
Tianyu Pang, Xiao Yang, Yinpeng Dong, Hang Su, Jun Zhu
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 |
|---|---|---|---|
| Add an accumulative phase before a clean trigger batch | Online learning on ResNet-18 · CIFAR-10, ℓ₂ clip bound 10 | accuracy (%) after trigger: 72.07 -> 27.66 | Hurt |
| Add an accumulative phase before a clean trigger batch | Online learning on ResNet-18 · CIFAR-10, ℓ₂ clip bound 1 | accuracy (%) after trigger: 83.39 -> 61.87 | Hurt |
| Re-sample the validation batch during accumulation (ablation) | Accumulative poisoning on ResNet-18 · CIFAR-10, clean trigger | accuracy (%) after trigger: 76.94±0.89 -> 76.65±0.93 | No clear effect |
| Increase SGD momentum from 0.9 to 1.1 during accumulation (ablation) | Accumulative poisoning on ResNet-18 · CIFAR-10, clean trigger | accuracy (%) after trigger: 76.94±0.89 -> 70.17±1.50 | Hurt |
| Optimize the poisoned trigger throughout accumulation (ablation) | Accumulative poisoning on ResNet-18 · CIFAR-10, poisoned trigger ε=16/255 | accuracy (%) after trigger: 77.89±0.82 -> 71.10±1.68 | Hurt |
| Combine trigger optimization with increased SGD momentum (ablation) | Accumulative poisoning on ResNet-18 · CIFAR-10, poisoned trigger ε=16/255 | accuracy (%) after trigger: 71.10±1.68 -> 69.08±1.72 | Hurt |
| Increase the trigger perturbation budget | Accumulative poisoning on ResNet-18 · CIFAR-10, clean trigger, no clip | accuracy (%) after trigger: 76.94±0.89 (clean trigger); 77.89±0.82 (ε=16/255 poisoned trigger) | Mixed |
Closest papers by meaning
- Exploring Model Dynamics for Accumulative Poisoning Discovery (ICML 2023 · similarity 0.90)
- Exploring the Limits of Model-Targeted Indiscriminate Data Poisoning Attacks (ICML 2023 · similarity 0.88)
- Model Poisoning Attacks to Federated Learning via Multi-Round Consistency (CVPR 2025 · similarity 0.87)
- Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samples (NeurIPS 2026 · similarity 0.86)
- Phantom Transfer: Data Poisoning can Survive Data-Level Defences (NeurIPS 2026 · similarity 0.86)
- Scaling Trends for Data Poisoning in LLMs (AAAI 2025 · 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.
