Research map / Adversarial robustness and security / Backdoor and poisoning
Anti-Backdoor Learning: Training Clean Models on Poisoned Data
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 |
|---|---|---|---|
| apply ABL training instead of standard training on poisoned data | WRN-16-1 · CIFAR-10, avg over 10 attacks | ASR (%): 97.73 -> 7.69 | Helped |
| use ABL instead of NAD defense | WRN-16-1 · CIFAR-10 clean accuracy | clean accuracy (%): 80.37 -> 84.76 | Helped |
| use ABL instead of NAD against Blend (CIFAR-10) and SIG (GTSRB) | WRN-16-1 · CIFAR-10 Blend | ASR (%): 4.56 -> 16.23 | Hurt |
| increase isolation rate p (ablation) | ABL · CIFAR-10, 6 classic attacks | ASR and clean accuracy | Mixed |
| delay turning epoch from 20 to 30 or 40 (ablation) | ABL · CIFAR-10 | ASR | Hurt |
| add local gradient ascent (gamma > 0) for isolation (ablation) | ABL · CIFAR-10 / ImageNet subset isolation precision | isolation precision | Helped |
| increase gamma from 0.5 to 1.5 (ablation) | LGA · CIFAR-10 BadNets | loss gap / stability | Mixed |
| use global gradient ascent unlearning instead of label shuffling (ablation) | ABL · CIFAR-10 BadNets | ASR (%): 30.23 -> 3.04 | Helped |
| use GGA instead of pixel noise on isolated examples (ablation) | ABL · CIFAR-10 BadNets | ASR (%): 57.54 -> 3.04 | Helped |
| use GGA instead of finetuning all layers on isolated clean set (ablation) | ABL · CIFAR-10 BadNets | ASR (%): 99.12 -> 3.04 | Helped |
| increase poisoning rate from 50% to 70% | ABL (1% isolation) · CIFAR-10 Dynamic | ASR (%): 25.74 -> 69.36 | Hurt |
Closest papers by meaning
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · similarity 0.92)
- BackdoorDM: A Comprehensive Benchmark for Backdoor Learning on Diffusion Model (NeurIPS 2025 · similarity 0.91)
- Demystifying Poisoning Backdoor Attacks from a Statistical Perspective (ICLR 2024 · similarity 0.91)
- Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks (ICLR 2025 · similarity 0.91)
- Mitigating Backdoor Attack by Injecting Proactive Defensive Backdoor (NeurIPS 2024 · similarity 0.91)
- From Toxic to Trustworthy: Using Self-Distillation and Semi-supervised Methods to Refine Neural Networks (AAAI 2024 · similarity 0.90)
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.