CUDA Autoencoder
A CIFAR-10 autoencoder with unified CPU and CUDA GPU pipelines, including naive GPU V1 and optimized GPU V2 kernels.
The problem
The project investigates how an autoencoder pipeline behaves when its computation is moved from CPU execution to CUDA kernels.
The approach
The repository provides C++ and CUDA implementations, GPU model code, kernel variants, run scripts, profiling notes, and training outputs.
Available materials
The public repository includes CUDA source, an autoencoder pipeline, CIFAR-10 data handling, profiling documents, notebooks, and reconstruction outputs.
Contribution & context
Author listed in the project README.
Scope of this project summary
The repository is a parallel programming final project tested on supported NVIDIA CUDA environments; no general production performance claim is made here.