This repo is a C++ implementation of various neurual networks(including feed forward, restricted Boltzmann machine, CNN, RNN, LSTM) for deep learning purpose. We use Armadillo C++ linear algebra library (http://arma.sourceforge.net/) to support our linear algebra operation in the CPU. We also have our GPU linear algebra library(wrapping cuBlas) to speed-up matrix/vector operations. Our implmentation supports seamless switch from CPU to GPU using a single Flag.