SNN对抗攻击笔记:
1. 解决SNN对抗攻击中脉冲与梯度数据格式不兼容性以及梯度消失问题:
- G2S Converter、Gradient Trigger[1]
2. 基于梯度的对抗攻击方式:
- FGSM、BIM[1]
3. 采用的神经元模型:
- 迭代LIF神经元模型[1]
4. 图像转化到脉冲序列的采样方式:
- Bernoulli采样[1]
- Poisson编码器[2]
5. 替代梯度法:
- 阶跃函数[1]
6. 影响对抗攻击效果的因素分析:
- 损失函数和发放阈值(倒数第二层)[1]
7.对抗攻击类型:
- 白盒攻击、目标/非目标攻击[1]
Reference:
[1] Liang L , Hu X , Deng L , et al. Exploring Adversarial Attack in Spiking Neural Networks with Spike-Compatible Gradient[J]. 2020.
[2] Sharmin S , Rathi N , Panda P , et al. Inherent Adversarial Robustness of Deep Spiking Neural Networks: Effects of Discrete Input Encoding and Non-Linear Activations[J]. 2020.
[3] Sharmin S , Panda P , Sarwar S S , et al. A Comprehensive Analysis on Adversarial Robustness of Spiking Neural Networks[J]. 2019.