Research map / Adversarial robustness and security / Certified and empirical robustness
Improved Gradient-Based Adversarial Attacks for Quantized Networks
Kartik Gupta, Thalaiyasingam Ajanthan
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 |
|---|---|---|---|
| Add temperature scaling with NJS to PGD | BNN-WQ ResNet-18 · CIFAR-10 | adversarial accuracy: 26.98 -> 0.00 | Improved |
| Add temperature scaling with HNS to PGD | BNN-WQ ResNet-18 · CIFAR-10 | adversarial accuracy: 26.98 -> 0.00 | Improved |
| Add temperature scaling with NJS to PGD | BNN-WAQ ResNet-18 · CIFAR-10 | adversarial accuracy: 8.57 -> 0.03 | Improved |
| Add temperature scaling with HNS to PGD | BNN-WAQ ResNet-18 · CIFAR-10 | adversarial accuracy: 8.57 -> 0.04 | Improved |
| Add temperature scaling with NJS to PGD | BNN-WQ ResNet-18 · CIFAR-100 | adversarial accuracy: 8.23 -> 2.45 | Improved |
| Add temperature scaling with HNS to PGD | BNN-WQ ResNet-18 · CIFAR-100 | adversarial accuracy: 8.23 -> 0.00 | Improved |
| Add temperature scaling to PGD | Adversarially trained floating-point ResNet-18 · CIFAR-10 | adversarial accuracy: 48.73 -> 47.17 (NJS); 48.73 -> 48.54 (HNS) | Mixed |
| Add temperature scaling to PGD | Adversarially trained BNN-WQ ResNet-18 · CIFAR-10 | adversarial accuracy: 41.34 -> 40.76 (NJS); 41.34 -> 40.67 (HNS) | Improved |
| Replace per-model temperature selection with fixed beta=0.1 (ablation) | PGD on adversarially trained floating-point ResNet-18 · CIFAR-10 | adversarial accuracy: 48.73 -> 61.27 | Hurt |
| Replace per-model temperature selection with fixed beta=0.1 (ablation) | FGSM on adversarially trained floating-point ResNet-18 · CIFAR-10 | adversarial accuracy: 62.38 -> 69.52 | Hurt |
| Use HNS rather than NJS for the one-step attack (ablation) | BNN-WQ ResNet-18 · CIFAR-10 | adversarial accuracy: FGSM++ NJS 3.46; FGSM++ HNS 2.51 | Improved |
| Increase PGD iterations, attack radius, or step size (ablation) | Original gradient-based attacks on BNNs · CIFAR-10 | adversarial accuracy | No clear effect |
Closest papers by meaning
- Defensive Quantization: When Efficiency Meets Robustness (ICLR 2019 · similarity 0.91)
- Quantization-Aware Interval Bound Propagation for Training Certifiably Robust Quantized Neural Networks (AAAI 2023 · similarity 0.89)
- Attacking Binarized Neural Networks (ICLR 2018 · similarity 0.89)
- TriQDef: Disrupting Semantic and Gradient Alignment to Prevent Adversarial Patch Transferability in Quantized Neural Networks (ICLR 2026 · similarity 0.88)
- Accurate, reliable and fast robustness evaluation (NeurIPS 2019 · similarity 0.88)
- $\sigma$-zero: Gradient-based Optimization of $\ell_0$-norm Adversarial Examples (ICLR 2025 · similarity 0.86)
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.
