Aij is a Java-based, low-dependency learning library which is designed to be exceptionally low level. I made this kit when I found that DeepLearning4j required too much effort to integrate into my project. While that library is significantly better in terms of speed and compatibility, Aij designed to let you throw a bunch of double arrays at a problem, get a trained RBM, and then move on with your life. There's no need to made complicated vectorization objects or subclass a bunch of different data iterators. Everything is plain and stupidly simple.
You can see it on github here: https://github.com/JosephCatrambone/Aij