Skip to main content

An Intro to Deep Learning with Apache MxNet Gluon

Dr. Steffen Rochel, AWS, Palo Alto, CA (USA)

07-Mai-2018, 17:00

Location: TU Berlin, MA 043

Abstract:

An Intro to Deep Learning with Apache MxNet Gluon

 

 

We will introduce Gluon, a flexible new interface that pairs MXNet’s speed with a user-friendly frontend. In the past, deep learning practitioners had to choose between ease of use and speed when choosing frameworks. On one side, there were symbolic frameworks like Theano and TensorFlow. These offer speed and memory efficiency but are harder to program, can be a pain to debug, and don’t allow for many native language features such as basic control flow. On the other side, there are imperative frameworks like Chainer and PyTorch. They’re a joy to program and easy to debug, but they can seldom compete with the symbolic code when it comes to speed. Gluon reconciles the two, removing a crucial pain point.
 Gluon can run as a fully imperative framework. In this mode, you enjoy native language features, painless debugging, and rapid prototyping. You can also effortlessly deploy arbitrarily complex models with dynamic graphs. But when you need more performance, Gluon can also provide the speed of MXNet’s symbolic API by calling down to Gluon’s just-in-time compiler. In this lecture, we’ll provide a short review of deep learning basics, the fundamentals of Gluon, advanced models, and multiple-GPU deployments. We'll show how to define neural networks through Gluon’s predefined layers. We’ll demonstrate how to serialize models and build dynamic graphs. Finally, we will show you how to hybridize your networks, simultaneously enjoying the benefits of imperative and symbolic deep learning.