Research map / Adversarial robustness and security / Attacks on text and graphs
BppAttack: Stealthy and Efficient Trojan Attacks Against Deep Neural Networks via Image Quantization and Contrastive Adversarial Learning
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 image quantization and dithering as an input-dependent trigger instead of WaNet's trigger | BppAttack · CIFAR-10, all-to-one | ASR: 99.55% -> 99.91% | Helped |
| Use image quantization and dithering as an input-dependent trigger instead of WaNet's trigger | BppAttack · CIFAR-10, all-to-all | ASR: 93.36% -> 94.32% | Helped |
| Use image quantization and dithering as an input-dependent trigger instead of WaNet's trigger | BppAttack · GTSRB, all-to-one | ASR: 98.78% -> 99.96% | Helped |
| Use image quantization and dithering as an input-dependent trigger instead of WaNet's trigger | BppAttack · CelebA, all-to-one | ASR: 99.33% -> 99.99% | Helped |
| Use image quantization and dithering as an input-dependent trigger instead of WaNet's trigger | BppAttack · MNIST, all-to-all | ASR: 95.90% -> 98.46% | Helped |
| Use image quantization and dithering as an input-dependent trigger instead of WaNet's trigger | BppAttack · ImageNet 200-class subset | ET: 450m -> 78m | Helped |
| Set quantization depth to 7 bits rather than 6 bits or fewer (ablation) | BppAttack · CIFAR-10, all-to-one | ASR | Hurt |
| Increase the poisoning injection rate (ablation) | BppAttack · CIFAR-10, all-to-one | ASR | Helped |
| Vary injection rate from 2.5% to 30% (ablation) | BppAttack · CIFAR-10, all-to-one | BA | No clear effect |
| Add Floyd–Steinberg dithering to quantization (ablation) | BppAttack · GTSRB trigger images | stealthiness | Helped |
| Replace contrastive adversarial training with vanilla training (ablation) | BppAttack · CIFAR-10 | Neural Cleanse anomaly index | Hurt |
| Add PGD-generated adversarial negative pairs to contrastive training (ablation) | BppAttack · CIFAR-10 | Neural Cleanse detection | Helped |
Closest papers by meaning
- Scanning Trojaned Models Using Out-of-Distribution Samples (NeurIPS 2024 · similarity 0.85)
- TrojanDec: Data-free Detection of Trojan Inputs in Self-supervised Learning (AAAI 2025 · similarity 0.84)
- Where the Devil Hides: Deepfake Detectors Can No Longer Be Trusted (CVPR 2025 · similarity 0.83)
- Attack To Defend: Exploiting Adversarial Attacks for Detecting Poisoned Models (CVPR 2024 · similarity 0.83)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · similarity 0.83)
- Data Poisoning based Backdoor Attacks to Contrastive Learning (CVPR 2024 · similarity 0.83)
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.