Amazon Web Services Inc. and Microsoft Corp., the two top players in the fiercely competitive cloud infrastructure market, make for unlikely allies. But recent history shows even the oldest industry rivalries can be set aside to pursue common interests.

Today, the two technology giants introduced a jointly developed machine learning library called Gluon. Available under an open-source license, the software is designed to take some of the complexity out of building deep learning neural networks, which are responsible for big advances in speech and image recognition and self-driving cars in recent years.

Large-scale AI projects rely on a large mix of technologies with an often steep learning curve to support the development workflow. Among the most central components is the training framework used to optimize the speed and accuracy of models. Developers typically choose one of the roughly half-dozen leading systems that are available in the open-source ecosystem.

These frameworks can make it quite difficult to program a neural network. Matt Wood, who heads artificial intelligence efforts at AWS, detailed in a blog post that developers must learn “arcane and rigid” methods of defining AI models that can require a lot of work.

“Developing with AI, especially deep learning models, isn’t easy—it can be a fairly daunting and specialized practice for most data professionals,” Eric Boyd, Microsoft corporate vice president for AI and infrastructure, added in a blog post. “This is another step in fostering an open AI ecosystem to accelerate innovation and democratization of AI—making it more accessible and valuable to all.”

The newly unveiled Gluon library masks some of the most tricky aspects behind a relatively simple programming interface. At a high level, the tool is intended to make building an AI model more like regular application development. Wood said that Gluon is easier to learn than the controls of many training frameworks and lets users implement features in a more concise way.

The latter feature reduces the amount of code required for a project, which makes it possible to roll out improvements faster when paired with Gluon’s simplified interface. This is further aided by the library’s support for what Wood called dynamic neural network graphs. According to the researcher, the feature enables users deploy an AI model in a form that can be modified more easily than regular implementations.

“Training of machine models is such a bottleneck in the development of useful machine learning models for specific tasks such as predicting some phenomenon of interest,” said Jim Kobielus, lead analyst for data science, deep learning and application development at Wikibon, the analyst group owned by the same company as SiliconANGLE. “What’s interesting about Gluon is it provides a mechanism for training of the model to be optimized on the fly to speed it up. This allows researchers to be more productive.”

AWS and Microsoft hope that streamlining development will make AI more accessible for companies. Gluon is not alone in offering features such as dynamic graphs to ease coding. But the tech giants said the library provides a much better balance between ease-of-use and performance than alternatives. 

Gluon is currently available as part of Apache MXNet, the open-source deep learning framework backed by AWS, with a future release set to add support for the Microsoft Cognitive Toolkit. The companies have also made it possible to integrate the library with other tools by releasing a reference specification.

Although it sounds unusual for the two cloud computing rivals to team up, they have a common enemy: Google. In 2015, Google, another leader in the development and use of machine learning, open-sourced the TensorFlow AI framework it developed, and it is now more popular than Cognitive Toolkit and MXNet.

With reporting from Robert Hof

Image: Clever Cogs!