This repository contains a pure CUDA C++ implementation of a Multilayer Perceptron (MLP) neural network with a Linear Regression Dataset with Apache 2 license on which the network attained a loss of 0.07. By building from the ground up, we gain a deep understanding of the inner workings of neural networks and the performance benefits of GPU acceleration.
- This project is for educational purposes and may not be optimized for production use.
- Pure CUDA C++ implementation: No external libraries or frameworks.
- GPU acceleration: Leverage the power of GPUs for high performance.
- Modular design: Clear separation of concerns for maintainability.
-
Clone the repository: Bash
git clone https://github.com/inventwithdean/CUDA_MLP.git -
Set up CUDA environment: Ensure you have a CUDA-capable GPU and the necessary CUDA toolkit installed.
-
Compile the code: Use a CUDA-compatible compiler to build the project.
-
Run the executable: Execute the generated binary to run the MLP.
- include: Header files for classes and functions along with some kernels for MSE and ReLUs.
- kernels: CUDA kernels for matrix related calculations
- main.cu: Main file where dataset is loaded and Optimization loop happens
- dataset.cu: A very simple and lightweight csv Dataset Loader class
- Implementing Softmax and Convolutional layers.
- Add optimization techniques like momentum.
- Explore different activation functions and network architectures.
- Improve performance through kernel tuning and optimization.
Feel free to contribute to this project by:
- Submitting bug reports
- Suggesting new features
- Improving the code
- Writing documentation
Let's dive into the world of deep learning together!
MIT