Research map / Adversarial robustness and security / Backdoor and poisoning
DEFEAT: Deep Hidden Feature Backdoor Attacks by Imperceptible Perturbation and Latent Representation Constraints
Check 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 DEFEAT instead of WaNet | ResNet34 on CIFAR-10 · CIFAR-10 | TAR: 92.03±0.30 -> 92.25±0.25 | Helped |
| Use DEFEAT instead of WaNet | ResNet34 on CIFAR-10 · CIFAR-10 | ASR: 99.96±0.01 -> 99.98±0.02 | No clear effect |
| Use DEFEAT instead of the strongest baseline | ResNet34 on GTSRB · GTSRB | TAR: 98.19±0.31 -> 98.26±0.30 | Helped |
| Use DEFEAT instead of BadNets | ResNet34 on GTSRB · GTSRB | ASR: 98.91±0.11 -> 99.01±0.05 | Helped |
| Use DEFEAT instead of BadNets | ResNet34 on ImageNet · ImageNet | TAR: 79.45±0.16 -> 82.63±0.22 | Helped |
| Use DEFEAT instead of DFST | ResNet34 on ImageNet · ImageNet | ASR: 98.59±0.03 -> 98.98±0.01 | Helped |
| Increase perturbation budget ϵ from 1 to 2 (ablation) | DEFEAT on CIFAR-10 · CIFAR-10 | TAR: 92.25±0.25 -> 92.65±0.22 | Helped |
| Increase perturbation budget ϵ from 1 to 2 (ablation) | DEFEAT on CIFAR-10 · CIFAR-10 | ASR: 99.98±0.02 -> 99.63±0.01 | Hurt |
| Increase perturbation budget ϵ from 2 to 3 (ablation) | DEFEAT on GTSRB · GTSRB | TAR: 98.26±0.30 -> 98.41±0.11 | Helped |
| Increase perturbation budget ϵ from 2 to 3 (ablation) | DEFEAT on GTSRB · GTSRB | ASR: 99.01±0.05 -> 99.61±0.01 | Helped |
| Increase perturbation budget ϵ from 3 to 5 (ablation) | DEFEAT on ImageNet · ImageNet | TAR: 82.63±0.22 -> 84.29±0.40 | Helped |
| Increase perturbation budget ϵ from 3 to 5 (ablation) | DEFEAT on ImageNet · ImageNet | ASR: 98.98±0.01 -> 99.91±0.03 | Helped |
Closest papers by meaning
- Backdoor Attack with Imperceptible Input and Latent Modification (NeurIPS 2021 · similarity 0.93)
- BAN: Detecting Backdoors Activated by Adversarial Neuron Noise (NeurIPS 2024 · similarity 0.89)
- A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives (AAAI 2024 · similarity 0.89)
- BaDExpert: Extracting Backdoor Functionality for Accurate Backdoor Input Detection (ICLR 2024 · similarity 0.89)
- Attention-Imperceptible Backdoor Attacks on Vision Transformers (AAAI 2025 · similarity 0.89)
- Inspecting Prediction Confidence for Detecting Black-Box Backdoor Attacks (AAAI 2024 · similarity 0.89)
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.