Research map / Adversarial robustness and security / Certified and empirical robustness
Inherent Adversarial Robustness of Deep Spiking Neural Networks: Effects of Discrete Input Encoding and Non-Linear Activations
Saima Sharmin, Nitin Rathi, Priyadarshini Panda, Kaushik Roy
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 |
|---|---|---|---|
| Train with surrogate-gradient backprop (SNN-BP) instead of ANN-SNN conversion only | SNN on VGG5 · CIFAR10 FGSM blackbox | adversarial accuracy: 16.0 vs 9.6 (PGD), 21.5 vs 19.3 (FGSM) | Improved |
| Use SNN-BP instead of ANN | VGG5 · CIFAR10 whitebox FGSM | adversarial accuracy: 10.4 -> 15 | Improved |
| Use SNN-BP instead of ANN | VGG11 · CIFAR100 whitebox FGSM | adversarial accuracy: 17.1 -> 15.5 | Hurt |
| Use SNN-BP instead of ANN | VGG11 · CIFAR100 whitebox PGD | adversarial accuracy: 8.5 -> 6.3 | Hurt |
| Use SNN-BP instead of ANN | VGG11 · CIFAR100 blackbox FGSM | adversarial accuracy: 21.2 -> 21.4 | No clear effect |
| Use SNN-BP instead of ANN | ResNet20 · CIFAR10 blackbox FGSM | adversarial accuracy: 56.7 -> 56.8 | No clear effect |
| Reduce number of timesteps from 150 to 50 (4-bit input quantization) (ablation) | SNN-BP VGG5 · CIFAR10 blackbox FGSM | adversarial accuracy: ~5% increase; clean 88.9 -> 87.5 | Improved |
| Add 2-bit input pre-quantization instead of 4-bit (ablation) | SNN-BP VGG5 · CIFAR10 blackbox FGSM | adversarial accuracy | Improved |
| Reduce number of timesteps (300 to 2000 range) (ablation) | SNN-conv VGG5 · CIFAR10 blackbox FGSM | adversarial accuracy | No clear effect |
| Add 2-bit input pre-quantization (ablation) | SNN-conv VGG5 · CIFAR10 blackbox FGSM | adversarial accuracy: ~21% to ~28% | Improved |
| Increase leak (decrease lambda to 0.95) (ablation) | SNN-BP VGG5 · CIFAR10 FGSM and PGD | adversarial accuracy: 3~6% increase | Improved |
| Replace ReLU with LIF neuron on perturbed input (ablation) | LIF neuron (lambda 0.99) · 32x32 image perturbation | l2 norm of perturbation: 0.6 -> 0.3 | Improved |
Closest papers by meaning
- Toward Robust Spiking Neural Network Against Adversarial Perturbation (NeurIPS 2022 · similarity 0.93)
- Enhancing Adversarial Robustness in SNNs with Sparse Gradients (ICML 2024 · similarity 0.93)
- Robust Stable Spiking Neural Networks (ICML 2024 · similarity 0.91)
- FEEL-SNN: Robust Spiking Neural Networks with Frequency Encoding and Evolutionary Leak Factor (NeurIPS 2024 · similarity 0.88)
- Towards Effective and Sparse Adversarial Attack on Spiking Neural Networks via Breaking Invisible Surrogate Gradients (CVPR 2025 · similarity 0.88)
- Enhancing the Robustness of Spiking Neural Networks with Stochastic Gating Mechanisms (AAAI 2024 · 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.
