Currently, the most popular DL libraries are TensorFlow (supported by Google), Keras (led by Francois Chollet, now at Google), and PyTorch (supported by Facebook). Development is very active, with PyTorch just having released version 1.0 and TensorFlow 2.0 expected in early Spring 2019, when it is expected to adopt Keras as its main interface.
All libraries provide the building blocks we discussed previously under Design choices, regularization and optimization algorithms, and facilitate fast training on Graphics Processing Units (GPUs). The libraries differ a bit in their focus with TensorFlow, which was originally designed for deployment in production, and Keras, which is more tailored for fast prototyping, although the interfaces are gradually converging.
We will illustrate the use of these libraries using the same network architecture and dataset as...
