Kolmogorov-Arnold Neural Networks (KAN)

This repository contains an implementation of Kolmogorov-Arnold Neural Networks (KAN) using PyTorch. The project demonstrates the application of KAN in classifying handwritten digits from the MNIST dataset.

Table of Contents

• Kolmogorov-Arnold Neural Networks (KAN)
  • Table of Contents
  • Introduction
  • Installation
  • Usage
  • Model Architecture
  • Training and Evaluation
  • Results

Introduction

Kolmogorov-Arnold Neural Networks (KAN) provide an alternative approach to traditional multi-layer perceptrons (MLPs) by leveraging the Kolmogorov-Arnold representation theorem. This project implements KAN and evaluates its performance on the MNIST dataset.

Installation

To get started, clone the repository and install the required dependencies:

git clone https://github.com/yourusername/kan.git
cd kan
pip install -r requirements.txt

Usage

You can run the provided Jupyter notebook to train and evaluate the Kolmogorov-Arnold Neural Network on the MNIST dataset.

1. Open the notebook kan.ipynb in Jupyter.
2. Execute the cells sequentially to train the model and evaluate its performance.

Model Architecture

The Kolmogorov-Arnold Neural Network (KAN) consists of multiple layers, each transforming input dimensions using linear layers and activation functions. The architecture is defined in the KolmogorovArnoldNN class in the notebook.

Training and Evaluation

The notebook covers the following steps:

1. Data Preparation: Loading and normalizing the MNIST dataset.
2. Model Definition: Creating an instance of the KolmogorovArnoldNN class.
3. Training: Training the model using the Adam optimizer and cross-entropy loss.
4. Evaluation: Calculating accuracy and visualizing results with a confusion matrix.

Results

The trained model's performance is evaluated on the test dataset, with results visualized using a confusion matrix. The confusion matrix provides insights into the model's accuracy across different classes.