Called Gluon, the deep learning library is intended to give all developers an interface that allows them to build machine learning models using a Python API and "a range of pre-built, optimized neural network components." Both Microsoft and AWS are hoping Gluon can help all developers, regardless of skill level, build neural networks in a simple way without impacting performance.
As Microsoft explains, deep learning engines like Apache MXNet, Microsoft Cognitive Toolkit, and TensorFlow have managed to speed up a training process that used to take days or weeks, but they require developers to implement lengthy and complex code to build training models. Other tools make the model building easier, but come with a slower training period. Gluon aims to strike a balance between the two, giving developers an easy programming interface and quick training. From Microsoft:
The Gluon interface gives developers the best of both worlds—a concise, easy-to-understand programming interface that enables developers to quickly prototype and experiment with neural network models, and a training method that has minimal impact on the speed of the underlying engine. Developers can use the Gluon interface to create neural networks on the fly, and to change their size and shape dynamically. In addition, because the Gluon interface brings together the training algorithm and the neural network model, developers can perform model training one step at a time. This means it is much easier to debug, update and reuse neural networks.
Artificial Intelligence (AI) and machine learning are set to become an even bigger part of the technology and apps that we use on a day-to-day basis, and Microsoft and Amazon are hoping that Gluon will give developers a leg up. The Gluon interface is open source and is available today in Apache MXNet 0.11. Support for Microsoft Cognitive Toolkit is scheduled to arrive in an upcoming release. You can find out how to get started with Gluon and MXNet with tutorials for beginners and experts, or check out the Gluon interface on Github.