Noisy SNN: Exploiting noise as a resource for computation and learning in spiking neural models
Date:
CSIG (China Society of Image and Graphics) “Neuromorphic Vision Seminar”
Noisy SNN: Exploiting noise as a resource for computation and learning in spiking neural models
BY Gehua Ma
This work presents a theoretical framework termed noisy spiking neural network (NSNN) to exploit the computational advantages originating from non-deterministic, noisy neural dynamics in the brain. Incorporating neuronal noise using NSNN improves the performance of spiking neural networks, demonstrates reliable and effective learning, and better reproduces the variability observed in biological neural computation. Because NSNN subsumes traditional deterministic spiking models, this model is advantageous for enabling more flexible and robust computation on neuromorphic hardware with inherent unreliabilities.