SpikingNerualNetwork_Project
This repo is a project aimed at learning and exploring the properties of spiking neural network (SNN) and comparing it with traditional neural network (refered as "artificial neural network (ANN)").
For the report for this project, see here.
What's different for SNN?
The idea of SNN come from Hebbian theory. Simply speaking, SNN enables each cell to generate "pulse", which can accumulate along time when corresponding cell is activated frequently. When the intense of such pulse reach certain "threshold", this cell will "fire" and backpropogation happens. On the other hand, if the pulse is not strong enough, the pulse intense (referred as "voltage" in biology field) will decay along time.
In such way, if we have lots of cells working together, cells with similar funtions will pulse together more frequently, this is where the saying "neurons that fire together wire together" comes from. This video also gives a really good view about how it works.
What's new in this repo?
Now the work is concentrating on the realization of SNN version of CNN network and Resnet module.
Reference
- This repo is based on a nice and clear repo of SNN to get quick start.
- Deep Residual Learning for Image Recognition