Enhancing the Robustness of Spiking Neural Networks with Stochastic Gating Mechanisms

Authors: Jianhao Ding, Zhaofei Yu, Tiejun Huang, Jian K. Liu

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results prove that our method can be used alone or with existing robust enhancement algorithms to improve SNN robustness and reduce SNN energy consumption.
Researcher Affiliation Academia 1School of Computer Science, Peking University 2Institution for Artificial Intelligence, Peking University 3School of Computer Science, University of Birmingham
Pseudocode No The paper describes its methods in prose and uses mathematical equations, but it does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https://github.com/Ding Jianhao/Sto G-meets-SNN/.
Open Datasets Yes To verify the effectiveness of our method, we conduct experiments on the CIFAR-10 and CIFAR100 datasets (Krizhevsky, Hinton et al. 2009).
Dataset Splits No The paper mentions using CIFAR-10 and CIFAR-100 datasets but does not explicitly state the training, validation, or test split percentages or sample counts within the main text.
Hardware Specification Yes The experiments are conducted on GPU devices of the NVIDIA RTX 3090 with PyTorch (v1.12.1).
Software Dependencies Yes The experiments are conducted on GPU devices of the NVIDIA RTX 3090 with PyTorch (v1.12.1).
Experiment Setup Yes To punish Po, we set γ = 5 10 6 by default. We train our model with white-box FGSM adversarial examples on each mini-batch of images. The perturbation boundary is 2/255 (Kundu, Pedram, and Beerel 2021). The EOT step is set to 10 by default... The intensity of the FGSM attack is 8/255. For the PGD-l attack, the overall intensity, step number, and step size are fixed to 8/255, 7, and 0.01, respectively.