Deep Learning Tools for IoT Are Released in an Open-Source Library

AWS and Microsoft have released Gluon, a library of deep learning code to help prototype AI faster. (Image courtesy of Amazon Web Services.)

Artificial intelligence is all the buzz these days. From Elon Musk shouting the sky will fall, to Mark Zuckerberg saying no bad will come from his Big Brother baby, there is no shortage of news on the AI topic.

Recently, Amazon Web Services (AWS) and Microsoft have brought us a few minutes closer to midnight on Musk’s doomsday clock with the release of their open-source deep learning interface Gluon. The aim of this new deep learning library is to allow users to quickly build machine learning models without compromising on training the AI system.

Gluon is designed to work with any deep learning engine and currently supports Apache MXNet. Soon, it will also support the Microsoft Cognitive Toolkit. The aim is for the library to give developers of any skill level the ability to create a machine learning model for the cloud, mobile apps and IoT devices on the edge. This sounds like it will fit perfectly with Internet of Things (IoT) development teams that don’t always have the experience they would like when it comes to designing their artificial intelligence (AI) systems.

The backbone to the Gluon interface is Python APIs and prebuilt neural network components. The result is the ability to use simple code to quickly produce AI without reducing performance. Gluon’s reference specification allows various deep learning engines to be integrated into its functionality.

Neural networks are built using training data, a model and an algorithm. The algorithm trains the model to pick up patterns in the data. This is a lengthy process, as both the model and algorithm are complex.

To shave days or weeks off of this process, deep learning engines like Apache MXNet, Microsoft Cognitive Toolkit and TensorFlow optimize the process. Unfortunately, this hasn’t reduced the complexity of the models and algorithms that become difficult and complex to write and change. There are deep learning tools that simplify the building of models, but Gluon developers report that there have been complaints about training speed and performance.

"I rarely see software engineering abstraction principles and numerical machine learning playing well together—and something that may look good in a tutorial could be hundreds of lines of code,” said Andrew Moore, dean of the School of Computer Science at Carnegie Mellon University. “I really appreciate how the Gluon interface is able to keep the code complexity at the same level as the concept; it’s a welcome addition to the machine learning community."

Gluon is designed to give IoT developers the ability to change their neural networks quickly. Because it also allows training and models to be altered in one interface, developers can start training at each step of the process. This helps with the debugging process as well as opens the door to update and reuse neural networks.

HAL 9000 Artificial Intelligence of 2001: A Space Odyssey. (Image courtesy of MGM.)

“We think the Gluon interface will be an important addition to our machine learning toolkit because it makes it easy to prototype machine learning models,” said Takero Ibuki, senior research engineer at DOCOMO Innovations. “The efficiency and flexibility this interface provides will enable our teams to be more agile and experiment in ways that would have required a prohibitive time investment in the past.”

You know I can’t help but relate to Musk’s AI fears. As AI becomes easy to develop, there will be no way to tell who will be playing around with it. For all you know, this article was written by an artificial intelligence to maximize your interest and likelihood of working with Gluon.

It wasn’t, but would I tell you the truth? What else will the AI monster be hiding once it learns everything about us from our iPhone, IoT devices, and Facebook and LinkedIn profiles?

I want to tell you. But “I’m sorry Dave, I’m afraid I can’t do that.”