LAS VEGAS — Amazon Web Services is making MXNet its deep-learning framework of choice, a top executive said this evening at the fifth annual re:Invent user’s conference. MXNet is an open-source product backed by multiple universities and companies, including Microsoft and the University of Washington.
“That means we’ll be providing code contributions, we’ll be investing in programmability and the developer experience, we’re going to be building up documentation and more example code, and we’ll invest in a series of tools around it,” said Matt Wood, AWS’s general manager of product strategies, during the evening’s 1.5-hour keynote. “In addition to that, MXNet is going to be the foundation of our future AI services.”
“Machine learning is going to be absolutely the dominant workload,” said James Hamilton, an AWS VP and Distinguished Engineer, during the keynote. “I believe it’s going to cause the growth I’ve shown you so far to look like small stuff. It’s going to be fundamentally important across the industry.”
MXNet allows defining, training and deploying deep neural networks on a wide array of devices, Wood said. It is easily programmed in any one of numerous popular languages.
In fact, it allows using both an imperative language, which is very flexible but hard to optimize, and a declarative language, which is easier to optimize but less flexible.
MXNet models are very compact, with a 1,000-layer neural network fitting in less than 4 GB, Wood said. That means they can be put inside drones or mobile devices. And MXNet is highly scalable, automatically parallellizing code and easily scaling up to run across multiple GPU instances.
Deep learning and neural networks lie underneath applications that distinguish images of cats from those of dogs and those that do natural-language processing.
In a blog post last week, Amazon CTO Werner Vogels explained why MXNet would be its “deep learning framework of choice,” noting how machine learning technologies are used at the company in autonomous drones, robotics in fulfillment centers, speech recognition and more.
Among machine learning algorithms, a class of algorithms called deep learning has come to represent those algorithms that can absorb huge volumes of data and learn elegant and useful patterns within that data: faces inside photos, the meaning of a text, or the intent of a spoken word. A set of programming models has emerged to help developers define and train AI models with deep learning; along with open source frameworks that put deep learning in the hands of mere mortals. Some examples of popular deep learning frameworks that we support on AWS include Caffe, CNTK, MXNet, TensorFlow, Theano, and Torch.
Among all these popular frameworks, we have concluded that MXNet is the most scalable framework. We believe that the AI community would benefit from putting more effort behind MXNet.
Matt Wood’s talk on deep learning begins at the 1:13:00 mark in the keynote video below.