CNN on CUDA

Implementation of Convolutional Neural Network using CUDA. On testing with MNIST dataset for 50 epochs, accuracy of 97.22% was obtained with a GPU training time of about 650 seconds.

Architecture

All tests performed on an Nvidia GeForce 840M GPU, running CUDA 8.0.61.

Compiling and Execution

To compile just navigate to root and type make Executable can be run using ./CNN