Parallel-Computing-GANs

A CUDA-accelerated project implementing core components of Generative Adversarial Networks (GANs) from scratch. This repository demonstrates how low-level parallel computing using NVIDIA CUDA and cuDNN can be used to implement high-performance deep learning modules.

---

Features

• Custom CUDA/C++ implementations of neural network layers
• Efficient memory management and GPU acceleration using CUDA
• Manual backpropagation logic for learning objectives
• Modular structure for experimentation and educational use

---

Directory Structure

Parallel-Computing-GANs/
├── include/
│   ├── activationLayer.cuh
│   ├── affineTransform.cuh
│   ├── batchNormalization.cuh
│   ├── convolution2D.cuh
│   ├── crossEntropy.cuh
│   ├── meanSquareError.cuh
│   └── utils.cuh
├── src/
│   ├── main.cu
│   └── testModules.cu
├── models/
│   └── ganNetwork.cuh
├── build/
│   └── [compiled binaries]
├── CMakeLists.txt
├── .clang-format
└── README.md

---

Implemented Modules

Module	Description
activationLayer.cuh	Implements ReLU, Sigmoid, Tanh functions with forward and backward support
affineTransform.cuh	Fully connected layer with weights, bias, and gradients
batchNormalization.cuh	CUDA-optimized batch normalization with mean/variance tracking
convolution2D.cuh	2D convolution layer with padding and stride options
crossEntropy.cuh	Cross-entropy loss computation with softmax activation
meanSquareError.cuh	Mean squared error for regression tasks
ganNetwork.cuh	Generator and Discriminator design using the above layers

---

Environment Setup

Dependencies

Ensure the following are installed:

• CUDA Toolkit (v11.0+):
  https://developer.nvidia.com/cuda-downloads

• cuDNN:
  https://developer.nvidia.com/cudnn-downloads

• CMake (v3.10 or newer)

---

Linux / WSL Setup

1. Export paths for CUDA and cuDNN:

   export PATH=/usr/local/cuda/bin:$PATH
   export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
   
   Here’s the finalized `README.md` content, properly compiled into a single markdown file. You can copy this directly into your GitHub repository's `README.md`. It includes all sections from your provided content, formatted cleanly and consistently:
   

---

2. Clone the repository:

   git clone https://github.com/codeclubiittp/Parallel-Computing-GANs.git
   cd Parallel-Computing-GANs

3. Build the project:

   mkdir build && cd build
   cmake ..
   make -j$(nproc)

---

Future Work

• Integrate pooling layers and dropout
• Implement complete training loop for GANs
• Support for real datasets like MNIST/CIFAR
• Tensor visualization for debugging

---