Research map / Adversarial robustness and security / Certified and empirical robustness
TriQDef: Disrupting Semantic and Gradient Alignment to Prevent Adversarial Patch Transferability in Quantized Neural Networks
Amira Guesmi, Bassem ouni, Muhammad Shafique
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 |
|---|---|---|---|
| use TriQDef instead of PBAT, 2-bit seen patches | TriQDef · CIFAR-10 LAVAN | ASR (%): 39.7 -> 26.2 | Improved |
| use TriQDef instead of PBAT on unseen patches, 32-bit | TriQDef · CIFAR-10 LAVAN | ASR (%): 77.2 -> 35.6 | Improved |
| use TriQDef instead of DWQ, ImageNet 2-bit | TriQDef · ImageNet LAVAN | ASR (%): 61.2 -> 28.5 | Improved |
| use TriQDef instead of Jedi, 2-bit QNN | TriQDef · ImageNet LAVAN (ResNet-50) | robust accuracy (%): 23.4 -> 65.8 | Improved |
| add FDP (ablation) | TriQDef · CIFAR-10 LAVAN, 2-bit | ASR (%): 55.9 -> 26.2 | Improved |
| add GPDP (ablation) | TriQDef · CIFAR-10 LAVAN, 2-bit | ASR (%): 37.6 -> 26.2 | Improved |
| add FDP (ablation) | TriQDef · ImageNet LAVAN, 2-bit | ASR (%): 52.1 -> 28.5 | Improved |
| add GPDP (ablation) | TriQDef · ImageNet LAVAN, 32-bit | ASR (%): 48.6 -> 35.0 | Improved |
| use TriQDef instead of standard QAT for clean accuracy, CIFAR-10 5-bit | TriQDef · CIFAR-10 clean accuracy | accuracy (%): 85.1 -> 83.3 | Hurt |
| use TriQDef instead of PBAT for clean accuracy, CIFAR-10 2-bit | TriQDef · CIFAR-10 clean accuracy | accuracy (%): 75.5 -> 75.8 | Improved |
Closest papers by meaning
- Improved Gradient-Based Adversarial Attacks for Quantized Networks (AAAI 2022 · similarity 0.88)
- Defensive Quantization: When Efficiency Meets Robustness (ICLR 2019 · similarity 0.87)
- Qu-ANTI-zation: Exploiting Quantization Artifacts for Achieving Adversarial Outcomes (NeurIPS 2021 · similarity 0.87)
- Durable Quantization Conditioned Misalignment Attack on Large Language Models (ICLR 2025 · similarity 0.87)
- Mind the Gap: A Practical Attack on GGUF Quantization (ICML 2025 · similarity 0.86)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 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.
