Amazon Web Services and Microsoft on Thursday announced the availability of Gluon, an open source deep learning library for building artificial intelligence neural networks.
Gluon will make it easier for researchers to define machine learning models using a collection of prebuilt, optimized neural network components. The interface will enable software developers and enterprise users to manipulate machine learning models like any other data structures.
Gluon also will enable data scientists and researchers to build prototypes quickly and utilize dynamic neural network graphs for entirely new model architectures without sacrificing training speed.
The Gluon interface allows developers of all skill levels to prototype, build, train and deploy sophisticated machine learning models for the cloud, devices at the edge, and mobile apps, according to both companies. The Gluon interface currently works with Apache MXNet and will support Microsoft Cognitive Toolkit in an upcoming release.
In providing the tool as open source, AWS and Microsoft published Gluon’s reference specification to better integrate other deep learning engines with the interface.
The Gluon interface is available now on GitHub.
“As AI tools become easier to use, more developers will create more apps infused with AI, which means AI will have a bigger impact on society. So tools like Gluon and Keras are helping to expand the impact of AI,” observed Chris Nicholson, CEO of Skymind .
Microsoft hopes the open source community will help to solve the enormous challenges confronting society. Developing with AI and deep learning models is a difficult endeavor for most data professionals, noted Eric Boyd, corporate vice president of AI data and infrastructure at Microsoft.
“We believe bringing AI advances to all developers, on any platform, using any language, with an open AI ecosystem, will help ensure AI is more accessible and valuable to all,” Boyd wrote in an online post.
A deep learning network is a large and complex operation that entails manually constructing millions of connections. Deep learning networks usually are unwieldy and difficult to debug, and their code often can not be reused between projects, according to Matt Wood, general manager of artificial intelligence at AWS.
Gluon is an easy-to-use Python application programming interface intended for non-AI specialists to prototype and train neural nets. A trained neural net is one that can produce accurate predictions about the data you feed it, Skymind’s Nicholson told LinuxInsider.
“So Gluon will help people experiment with neural net configurations. What they do not do is offer a production-grade AI server. They are stuck in the training phase and do not do deployment like Skymind’s tools,” he said.
What Gluon Does
Gluon brings together two key components: training algorithms and neural network models. It eases the effort of integrating AI with applications, said Mark Lambiase, chief technology officer of Fox Technologies.
“I believe the play here is the vast quantities of data necessary to have AI train to a level of relevancy. Big data has opened the way to store more types of data intelligently,” he told LinuxInsider.
Big data relies on massive amounts of computers sorting, indexing and storing the data. The storage necessary for a good AI will eclipse the infrastructure necessary for the algorithms that will scan the data to make sense of it, Lambiase explained.
Gluon has lowered the bar for adding AI to an application, he added. Developers are going to be able to use this to start adding smarts to their applications.
“This can potentially open up a whole new industry for data brokers who could … sell subscriptions to their data to application developers that employ AI but do not have the resources to collect the data on their own,” Lambiase said.
Open Source Critical
Both Amazon and Microsoft have been battling Google to build the dominant AI framework. Right now, Google is winning, noted Nicholson.
Gluon competes with Keras, an AI framework tool that integrates with Google’s TensorFlow. Both Keras and TensorFlow were built by Google engineers.
“Until recently, Keras looked like it would be a standard API over multiple frameworks like Microsoft’s CNTK and Amazon’s MxNet. But with Gluon, it is clear that Microsoft and Amazon want to limit the reach of Google’s tech,” said Nicholson.
AWS and Microsoft are aware that today’s developer communities are forming around open source software. Releasing Gluon as open source is another data point reflecting that reality, said Lucas Geiger, CEO of Wireline.
The open source donation proves “Microsoft has awakened to the reality that it is the open source developers that have the strongest communities, and those communities are also influencing business decisions,” he told LinuxInsider.
Battle of Influence
One of the benefits of open sourcing AI is to get it into the hands of developers who can seed it into every application and service available, according to Lambiase.
“With AI, the genie is out of the bottle, and trying to monopolize it will more likely alienate people than win any friends … . AI itself is not going to drive revenue directly for these companies,” he said.
Another benefit to both AWS and Microsoft in open-sourcing the shared Gluon project is that they can cast a wider net to bring a diverse set of talents to the community, Wireline’s Geiger suggested. “Additionally, many of the top tier talent would distrust any efforts in AI that were not open sourced. That is the new reality of software development. Open source is the only viable path.”