Deep learning is a rapidly advancing domain in artificial intelligence, and Deeplearning4j is the most widely used open-source deep learning framework for the JVM. http://deeplearning4j.org/
It includes recurrent neural networks for time series and sound, word2vec and other algorithms for text, and convolutional networks for images.
Our distributed, deep artificial neural networks are powered by ND4J, a scientific computing library that supports n-dimensional arrays on the JVM. To build ND4J, we created a DSL that ported Numpy to the JVM. Under the hood, ND4J is optimized in C++. http://nd4j.org/
We built DL4J and ND4J to handle very large datasets, integrating them with Hadoop and Spark to train neural nets in a massively parallel fashion. In the process of optimizing our system, we wrote our own CUDA kernels and overhauled parts of the JVM.
The use cases deep learning can address include recommender systems, image processing and fraud detection.
Please check out our code on Github to learn more!