I have been using Neural Networks since the 1980s, and I just finished a job at Capital One managing a deep learning team. I work mostly with GANs and RNN.
HLearn looked good, but I didn’t spend much time with it. I suggest you try both HLearn and the Haskell TensorFlow bindings. If you don’t try to set up using a GPU then the setup is fairly easy. The Haskell TensorFlow minst example is very cool with terminal graphics, but I was unable to get an RNN working.
