When it comes to solving a problem the deep learning perspective is very different from the standard development approach. Instead of designing an algorithm to achieve a solution you find yourself looking at the data (preferably Big Data), trying to find out a good way to normalize them and to represent them, both as input and as output, in an effective way to the neural network.
Eclipse Deeplearning4j (DL4J) allows you to get into the play quicky, leveraging on your Java background and skills. It includes several libraries and APIs for data normalization, designing, training, testing, storing and reloading a network. Further it integrates with other popular deep learning Python tools like Keras and TensorFlow.