Research map / Adversarial robustness and security / Attacks on text and graphs
HIRE-SNN: Harnessing the Inherent Robustness of Energy-Efficient Deep Spiking Neural Networks by Training With Crafted Input Noise
Souvik Kundu, Massoud Pedram, Peter A. Beerel
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 crafted-noise training across time-step periods (HIRE-SNN) to traditional SNN training | direct-input SNN · CIFAR-10 white-box | accuracy under attack: up to 12.2% improvement over traditional SNN training | Improved |
| add crafted-noise training (HIRE-SNN) to traditional SNN training | direct-input SNN · CIFAR-100 white-box | accuracy under attack: up to 8.8% | Improved |
| add crafted-noise training (HIRE-SNN) to traditional SNN training | direct-input SNN · CIFAR-10 black-box | accuracy under attack: up to 13.7% | Improved |
| add crafted-noise training (HIRE-SNN) to traditional SNN training | direct-input SNN · CIFAR-100 black-box | accuracy under attack: up to 10.4% | Improved |
| use crafted gradient-based noise instead of Gaussian noise in training (ablation) | HIRE-SNN VGG11 · CIFAR-100 white-box | accuracy under attack: up to 6.5% | Improved |
| use crafted gradient-based noise instead of Gaussian noise in training (ablation) | HIRE-SNN VGG11 · CIFAR-100 black-box | accuracy under attack: up to 9.7% | Improved |
| add Gaussian noise training in place of traditional training (ablation) | SNN VGG11 · CIFAR-100 white-box and black-box | accuracy under attack | No clear effect |
| add crafted-noise training when threshold and leak are frozen (ablation) | HIRE-SNN VGG5 · CIFAR-10 white-box PGD | accuracy under attack: 4.5 -> 13.6 | Improved |
| add crafted-noise training when threshold and leak are frozen (ablation) | HIRE-SNN VGG5 · CIFAR-10 black-box PGD | accuracy under attack: 8.8 -> 21.9 | Improved |
| make threshold and leak trainable (ablation) | HIRE-SNN VGG5 · CIFAR-10 clean | accuracy: 86.8 -> 87.5 | Improved |
| increase noise step eps_s from 0.01 to 0.03 (ablation) | HIRE-SNN VGG5 and VGG11 · CIFAR-10 and CIFAR-100 | adversarial vs clean accuracy: drop in clean image performance of up to ~2% | Mixed |
| add period-wise weight update with noise crafting (N=2) to traditional training | HIRE-SNN · VGG5/VGG11/ResNet12 training cost | GPU memory: up to ~25% less memory, no extra training time | Improved |
Closest papers by meaning
- Enhancing Adversarial Robustness in SNNs with Sparse Gradients (ICML 2024 · similarity 0.92)
- S$^2$NN: Sub-bit Spiking Neural Networks (NeurIPS 2025 · similarity 0.90)
- EnOF-SNN: Training Accurate Spiking Neural Networks via Enhancing the Output Feature (NeurIPS 2024 · similarity 0.90)
- Robust Stable Spiking Neural Networks (ICML 2024 · similarity 0.89)
- Enhancing Training of Spiking Neural Network with Stochastic Latency (AAAI 2024 · similarity 0.89)
- FEEL-SNN: Robust Spiking Neural Networks with Frequency Encoding and Evolutionary Leak Factor (NeurIPS 2024 · similarity 0.89)
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.
