Research map / Adversarial robustness and security / Backdoor and poisoning
DIFT: Protecting Contrastive Learning Against Data Poisoning Backdoor Attacks
Jiang Zhu, Yulin Jin, Qingqing Ye, Zhibiao Guo, Kun Fang, Ruochen Du, Yingnan Zhao, Haibo Hu
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 |
|---|---|---|---|
| apply DIFT in-training defense | SimCLR · CIFAR-10 (CTRL) | ASR (%): 99.0 -> 2.0 | Improved |
| apply DIFT instead of post-training ensemble defense (B2) | SimCLR · CIFAR-10 (CTRL) | ASR w/ defense (%): 13.1 -> 2.0 | Improved |
| apply DIFT on ImageNet-100 | SimCLR · ImageNet-100 (CTRL) | ASR w/ defense (%): 49.1 -> 5.3 | Improved |
| apply DIFT instead of retraining baseline on accuracy | SimCLR · CIFAR-10 (CTRL) | accuracy drop (%): B2 1.4 -> DIFT 0.4 | Improved |
| add fine-tuning on detected poisoned samples with density-reduction loss (ablation) | DIFT · CIFAR-10 (SimCLR) | ASR w/ defense (%): 86.4% -> 2.0 | Improved |
| add label spreading after density-based detection (ablation) | DIFT · CIFAR-10 and CIFAR-100 | recall of poisoned samples | Improved |
| add label spreading after density-based detection (ablation) | DIFT · CIFAR-10 and CIFAR-100 | precision of poisoned sample detection | Hurt |
| omit label spreading from DIFT (ablation) | DIFT · final model | residual ASR | Hurt |
Closest papers by meaning
- Data Poisoning based Backdoor Attacks to Contrastive Learning (CVPR 2024 · similarity 0.90)
- Backdoor Contrastive Learning via Bi-level Trigger Optimization (ICLR 2024 · similarity 0.88)
- Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks (ICLR 2025 · similarity 0.86)
- BackdoorDM: A Comprehensive Benchmark for Backdoor Learning on Diffusion Model (NeurIPS 2025 · similarity 0.86)
- RPP: A Certified Poisoned-Sample Detection Framework for Backdoor Attacks under Dataset Imbalance (NeurIPS 2026 · similarity 0.86)
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 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.
