GPU Optimization for Shallow Neural Network Training using CUDA

A neural-network training project implemented in C, pthreads, CUDA.

Requirements

• GCC with OpenMP support.
• NVIDIA CUDA Toolkit (nvcc) and a CUDA-capable GPU for CUDA versions.

Project layout

• refrence/ — sequential, pthreads, and baseline CUDA implementations.
• alternative/ — optimized CUDA implementation.
• refrence/data/ — CSV datasets.
• project_paper.pdf — project report.

Run

Run the commands from the indicated directory.

# Sequential C version
cd refrence
bash scripts/run.sh

# Pthreads version
bash scripts/run_pthreads.sh

# CUDA version
bash scripts/run_cuda.sh

# Optimized CUDA version
cd ../alternative
bash run_cuda_optimized.sh

Each script builds its program and trains it using synthetic_convex_large.csv.