Neuromorphic_Computing

Overview

Neuromorphic computing is a hardware and software paradigm inspired by the structure of the human brain. This project benchmarks two neural network approaches on the MNIST handwritten-digit dataset:

• SNN (Spiking Neural Network) — a biologically inspired model that communicates via discrete spikes, implemented with torch, torchvision, and spikingjelly.
• CNN (Convolutional Neural Network) — a standard deep-learning model, implemented with keras (TensorFlow backend), numpy, and scikit-learn. The goal is to compare the accuracy and run-time efficiency of SNNs vs CNNs for empirical research purposes.

Requirements

• Python 3.8 or higher

Installation

1. Clone the repository:

   git clone https://github.com/RadoKyselak/Neuromorphic_Comp.git
   cd Neuromorphic_Comp

2. Create a virtual environment and activate it:

   python -m venv venv
   venv\Scripts\activate  # On Mac use `source venv/bin/activate`

3. Install the required packages:

   pip install -r requirements.txt

Usage

To run the SNN benchmark:

python "SNN benchmark.py"

To run the CNN benchmark:

python "CNN benchmark.py"

Both scripts automatically download the MNIST dataset on first run. The SNN script prints the start/end time and training loss, then reports test accuracy. The CNN script trains for 4 epochs and reports final accuracy.

License

This project is licensed under the Apache License. See the LICENSE file for details.

Contact

If you have any questions or feedback, please open an issue.