MNIST Simple Neural Network

A simple implementation of a neural network for the MNIST handwritten digit classification task using PyTorch.

Project Description

This project implements a basic neural network architecture to classify handwritten digits from the MNIST dataset. The network features:

• Two linear layers with ReLU activation in between
• Cross-entropy loss function with softmax
• PyTorch implementation

Architecture

The neural network consists of:

1. Input layer (784 neurons - flattened 28x28 MNIST images)
2. First linear layer with ReLU activation
3. Second linear layer
4. Softmax layer (part of cross-entropy loss)

Requirements

• Python 3.x
• PyTorch
• torchvision
• numpy

Usage

The project is implemented in a Jupyter notebook (MNIST_Simple_NN.ipynb). To run the project:

1. Install the required dependencies
2. Open the Jupyter notebook
3. Run the cells sequentially to:
   • Load and preprocess the MNIST dataset
   • Create and train the neural network
   • Evaluate the model's performance

Results

The model demonstrates the basics of deep learning by achieving reasonable accuracy on the MNIST dataset, serving as a good starting point for understanding neural networks and PyTorch implementation.

License

This project is open source and available under the MIT License.