Research map / Adversarial robustness and security / Attacks on text and graphs
Fine-Grained Iterative Adversarial Attacks with Limited Computation Budget
Zhichao Hou, Weizhi Gao, Xiaorui Liu
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 |
|---|---|---|---|
| replace full-precision attack with Spiking-PGD in adversarial training (lambda=2.0) | PGD-AT (ResNet-18) · CIFAR-10 | clean / robust accuracy: clean 80.18 -> 82.33; robust 48.75 -> 47.45 | Mixed |
| replace full-precision attack with Spiking-PGD in adversarial training (lambda=3.0) | PGD-AT (ResNet-18) · CIFAR-10 | robust accuracy: 48.75 -> 48.88 | No clear effect |
| use exponential decay threshold schedule instead of constant threshold (ablation) | Spiking-PGD-AT · CIFAR-10 | robust accuracy: constant rho=0.05: 37.59 vs lambda=3.0: 48.88 | Improved |
| use larger constant spiking threshold (rho 0.01 -> 0.05) (ablation) | Spiking-PGD-AT · CIFAR-10 | robust accuracy: 49.04 -> 37.59 | Hurt |
| use smaller decay rate lambda=1.0 instead of 3.0 (ablation) | Spiking-PGD-AT · CIFAR-10 | robust accuracy: 48.88 -> 43.66 | Hurt |
| add spiking forward computation at equal cost | PGD / I-FGSM / MI-FGSM · CIFAR-10, CIFAR-100, Tiny-ImageNet | accuracy under attack | Improved |
| add spiking forward computation at equal cost | PGD topology attack (GCN) · Cora, Citeseer | accuracy under attack | Improved |
| add virtual surrogate gradient (ablation) | Spiking-PGD · CIFAR-10 | attack strength | Improved |
| raise spiking threshold rho (ablation) | Spiking-PGD · CIFAR-10 | attack strength vs cost | Mixed |
Closest papers by meaning
- A Simple Class-Agnostic Approach to Enhance Fair Adversarial Training (NeurIPS 2026 · similarity 0.87)
- Efficient Robustness Evaluation via Constraint Relaxation (AAAI 2025 · similarity 0.87)
- Scaling Laws for Adversarial Attacks on Language Model Activations and Tokens (ICLR 2025 · similarity 0.86)
- Statistical Estimation of Adversarial Risk in Large Language Models under Best-of-N Sampling (NeurIPS 2026 · similarity 0.86)
- AttackBench: Evaluating Gradient-based Attacks for Adversarial Examples (AAAI 2025 · similarity 0.86)
- SuperDeepFool: a new fast and accurate minimal adversarial attack (NeurIPS 2024 · 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.
