Research map / Adversarial robustness and security / Attacks on text and graphs
On the Role of Temporal Granularity in the Robustness of Spiking Neural Networks
Mengting Xu, Shi Gu, Peng Lin, De Ma, Huajin Tang, Qian Zheng, Gang Pan
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 |
|---|---|---|---|
| attack with per-time-step gradients (TG-FGSM) instead of time-averaged gradient | vanilla VGG11 SNN · CIFAR10, T=4 | accuracy under FGSM: 12.71 -> 9.72 | Improved |
| add TG-Reg to standard training | VGG11 SNN · CIFAR10, T=4, PGD | accuracy under attack: 0.38 -> 4.25 | Improved |
| add TG-Reg to standard training | VGG11 SNN · CIFAR10, T=4, FGSM | accuracy under attack: 12.71 -> 28.79 | Improved |
| add TG-Reg to standard training | ResNet34 SNN · CIFAR10, T=4, clean | clean accuracy: 93.17 -> 87.32 | Hurt |
| add TG-Reg to standard training | VGG11 SNN · CIFAR100, T=8, clean | clean accuracy: 72.76 -> 70.98 | Hurt |
| add TG-Reg on top of PGD adversarial training | VGG11 SNN · CIFAR10, T=4, PGD | accuracy under attack: 22.35 -> 31.12 (vs RAT) | Improved |
| use TG-Reg instead of FEEL-SNN | VGG11 SNN · CIFAR100, T=8, PGD | accuracy under attack: 2.04 -> 4.59 | Improved |
| add TG-Reg | VGG11 SNN · CIFAR100, T=8, black-box PGD | accuracy under attack: 5.77 -> 35.91 | Improved |
| add TG-Reg on DVS-CIFAR10 | VGG11 SNN · DVS-CIFAR10, T=10, PGD | accuracy under attack: 11.20 -> 31.70 | Improved |
| add TG-Reg to WideResNet16 | WideResNet16 SNN · CIFAR100, T=8, PGD | accuracy under attack: 0.02 -> 0.33 | No clear effect |
| add TG-Reg to Tiny-ImageNet ResNet19 | ResNet19 SNN · Tiny-ImageNet, T=4, PGD | accuracy under attack: 2.91 -> 3.87 | Improved |
| add TG-Reg regularization (training cost) | VGG11 SNN · CIFAR100, T=8 | training time per epoch: 108s -> 236s | Hurt |
Closest papers by meaning
- Enhancing Adversarial Robustness in SNNs with Sparse Gradients (ICML 2024 · similarity 0.90)
- Training High Performance Spiking Neural Network by Temporal Model Calibration (ICML 2025 · similarity 0.89)
- Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness (ICLR 2025 · similarity 0.89)
- Robust Stable Spiking Neural Networks (ICML 2024 · similarity 0.88)
- Rethinking Spiking Neural Networks from an Ensemble Learning Perspective (ICLR 2025 · similarity 0.87)
- TS-SNN: Temporal Shift Module for Spiking Neural Networks (ICML 2025 · similarity 0.87)
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.
