NeuroBender: Enhanced Image Processing with K-A Networks & WAE

Welcome to NeuroBender, where the mystical powers of mathematical complexity meet the robust stability of autoencoders! This project combines the Kolmogorov-Arnold network with a Wasserstein Auto-Encoder to provide a unique approach to image processing on the MNIST dataset. Dive into a world where digits are not just recognized but transformed and reconstructed in new and interesting ways!

Project Overview

NeuroBender takes the classical MNIST dataset and applies a two-stage process:

1. Transformation using the Kolmogorov-Arnold Network: This step involves a complex transformation of the image data to enhance features.
2. Reconstruction using the Wasserstein Auto-Encoder: Post-transformation, the data is fed into a WAE to reduce dimensionality and reconstruct the images.

This approach is designed to test the synergy between complex mathematical transformations and modern deep learning techniques.

Getting Started

Follow these instructions to get a copy of the project up and running on your local machine for development and testing purposes.

Prerequisites

You'll need Python 3.8 or later, and the following packages:

• TensorFlow
• NumPy
• Matplotlib

Installation

1. Clone the repository:

   git clone https://github.com/JaroslawHryszko/neurobender.git

2. Navigate to the project directory:

   cd neurobender

3. Install required Python packages:

   pip install -r requirements.txt

Running the Project

To run NeuroBender and see it in action:

python integrated_model.py

This will execute the process on a subset of MNIST data and display the original and reconstructed images.

Code Structure

• kolmogorov_arnold_network.py: Contains the implementation of the Kolmogorov-Arnold network.
• wae_mnist.py: Houses the Wasserstein Auto-Encoder setup.
• integrated_model.py: Integrates both models and includes preprocessing and prediction logic.
• test_integrated_model.py: Contains unit tests for verifying the functionality of integrated systems.