Nearly every major tech company, from Google Inc. to Facebook Inc., has been making deep learning an increasingly important aspect of their businesses, and today Amazon.com Inc. got into the game.

The company, whose Amazon Web Services is the leader in cloud computing, announced that it has chosen the MXNet software framework on which it will build its own foundation for deep learning, a branch of artificial intelligence that attempts to emulate in software the way the brain learns.

Although MXNet will be future of deep learning at Amazon, the company has already been using deep learning to power a wide range of services for some time now.

“At Amazon, machine learning has been key to many of our business processes, from recommendations to fraud detection, from inventory levels to book classification to abusive review detection,” Amazon Chief Technology Officer Werner Vogels (above) explained in a blog post. “And there are many more application areas where we use machine learning extensively: search, autonomous drones, robotics in fulfillment centers, text and speech recognitions, etc.”

Vogels said that Amazon tested and considered a number of deep learning frameworks, including Google’s TensorFlow, but the company ultimately decided on MXNet due to its scalability. “We believe that the AI community would benefit from putting more effort behind MXNet,” he said.

Amazon Web Services will be contributing code and new documentation to MXNet, and while Vogel says that the company will continue investing in several different deep learning technologies, he noted that “we plan to contribute significantly to one in particular, MXNet.”

According to Vogels, Amazon evaluated different frameworks based on three key criteria:

1. Ability to scale to multiple GPUs: This allows Amazon to take advantage of massive amounts of compute power to deal with especially complicated machine learning tasks.
2. Development speed and programmability: Essentially, Amazon wanted a framework that its developers could pick up quickly and start using right away with many of the same skills they already have.
3. Portability: Machine learning that is limited to big server farms has limited uses, and Amazon wants to be able to use these tools on a wide range of devices.

Vogels explained that scalability is “one of [MXNet’s] defining features,” and it supports a number of common programming languages known by many developers, including Python, C++ and Javascript. He also noted that MXNet has an incredibly small footprint, requiring a minimum of only 4 GB of memory, and it can be ported across multiple operating systems, including both Android and iOS.

“It’s still day one for this new era of machine intelligence; in fact, we probably haven’t even woken up and had our first cup of coffee yet,” Vogels concluded. “With tools like MXNet (and the other deep learning frameworks), and services such as EC2, it’s going to be an exciting time.”

Photo by Guido van Nispen (http://www.flickr.com/photos/vannispen/3012063215) [CC BY-SA 2.0], via Wikimedia Commons