Build CUDA Neural Network From Scratch

This project is a CUDA-based neural network implementation, developed from scratch with performance optimizations and modifications. The project is inspired by cuda-neural-network and has been extended with additional features and optimizations to improve performance. Notably, the input dataset has been changed to MNIST, a popular dataset for handwritten digit recognition.

🚀a c++ version is here https://github.com/Phoenix8215/build_neural_network_from_scratch_CPP

Features

• Fully implemented neural network in CUDA for high-performance computations.
• Optimized CUDA kernels for enhanced performance.
• MNIST dataset integration for training and testing the neural network.
• Customizable network architecture to experiment with different configurations.

Installation

Prerequisites

• CUDA: Ensure that you have CUDA installed on your system. You can download it from NVIDIA's website.
• Make: Required for building the project.
• A modern C++ compiler: GCC or Clang is recommended.

Steps

1. Clone the repository:

   git clone https://github.com/Phoenix8215/BuildCudaNeuralNetworkFromScratch
   cd BuildCudaNeuralNetworkFromScratch

2. Build the project using Make:

   make

3. Run the application:

   ./cuda-conv

Contributing

Contributions are welcome! If you'd like to contribute, please fork the repository and make your changes in a separate branch. Then, submit a pull request with a description of your changes.

Acknowledgments

• Thanks to pwlnk for the original cuda-neural-network project, which served as the inspiration and foundation for this work.
• Thanks to the developers and contributors of the CUDA toolkit and the MNIST dataset.