Noise Robustness in SNNs
Spiking neural networks tolerate synaptic noise far better than ANNs thanks to temporal spike integration acting as a natural noise filter.
Spiking neural networks (SNNs) process information through discrete spikes accumulated over time. This temporal integration acts as a natural noise filter: the signal-to-noise ratio increases with the number of spikes per synapse, making SNNs inherently more robust to synaptic weight noise than ANNs.
Key Results:
- SNN shows <0.2% accuracy drop vs ANN shows 75% accuracy drop (CNN on MNIST; noise level = 100% x max weight)
Publication: Li C, Chen R, Moutafis C, Furber S. Robustness to Noisy Synaptic Weights in Spiking Neural Networks. IJCNN 2020.