izhinet

This repository contains a pure numpy based implementation of Izhikevich neurons as a spiking neural network. It is designed for clarity and simplicity, for complex simulations with larger network sizes have a look at Brian2.

Getting Started

Everything required to run a simulation and plot the spike trains are provided in izhinet.py with example code in run.py showing how to run. To install the dependecies you can:

pip3 install --no-cache-dir --upgrade -r requirements.txt

If you are Anaconda or other Python environment, you'll to install numpy and matplotlib following their instructions. Once you have the dependencies setup:

python3 run.py -h
  usage: run.py [-h] [-rt RUNTIME] [-dt DELTAT]

  Run random spiking neural networks.

  optional arguments:
    -h, --help            show this help message and exit
    -rt RUNTIME, --runtime RUNTIME
                          Simulation runtime in milliseconds per input.
    -dt DELTAT, --deltat DELTAT
                          Simulation delta time (dt), resolution.

For details on the simulation, you can refer to the paper and the code in izhinet.py for the actual implementation.

Limitations & To-Dos

• Currently only the spikes / firings are stored, the state variables v and u can also be tracked albeit with extra memory usage.
• The input current is fixed for the duration of the run, a timed input - one that changes at certain invervals - can be implemented to provide more flexibility.
• It seems numpy runs on a single core, moving to a multi-threaded numerical computation library might be worth exploring.

Built With

• NumPy
• matplotlib