Close
Close

Published

AWS, Microsoft, hope Gluon AI library will stick for machine-learning

Amazon’s AWS and Microsoft have launched Gluon, a new open source deep-learning interface that the pair hope will allow developers to more easily build machine-learning computing models – and boost demand for AWS and Azure’s cloud computing services. The pair have collaborated on entwining their Cortana and Alexa digital assistants in the past, so aren’t strangers to AI-based cooperation, but are firm rivals in the cloud computing market.

So now that the introduction is out of the way, we can note that it is still very early days for the AI and ML markets – with the developer ecosystems still taking shape, and no clear standard approach taking the lead. There are still dozens of ways of doing roughly the same thing, and what we are seeing is the major industry stakeholders planting their own AI and ML stakes in the ground – hoping to become influential early, and reap the rewards later on.

These machine-learning functions are set to be incredibly valuable in the IoT, due to the vast amount of data that all the extra IoT devices and links is expected to generate. There aren’t enough man-hours in the world to manually inspect this data, and existing automated systems aren’t ‘smart’ enough to make actual sense of the diverse sets of data – only understanding slices, not the picture as a whole.

As such, machine-learning can help automate the data processing, and the specific functions or applications can be combined into an application that one would not be amiss calling ‘intelligent,’ – in so far as it being able to adapt to new circumstances on the fly, and act independently of human intervention. That sort of system has huge value to the large business software providers (the likes of SAP, IBM, Cisco, Microsoft, and AWS), but we’re a few years away from those becoming prevalent.

For AWS and Microsoft in particular, they hope that opening the doors to AI and ML to other developers, via Gluon, will go a long way to boosting demand for their products. After all, Gluon is open source, and so they can’t charge a license for it. Instead, they hope that new applications will be built, which will then need the cloud compute power to run them – which AWS and Microsoft can then sell to them.

So while the two will be competing for those contracts, ready to spool up cloud instances when needed, they have a mutual benefit to work towards using Gluon – namely, simplifying and streamlining the ecosystem, so that they don’t have to support dozens of different implementations. Having three dominant ML approaches instead of 10-20 benefits everyone trying to cater for a mass audience – but there will still be plenty of room for super-niche approaches from smaller players.

We believe it is important for the industry to work together and pool resources to build technology that benefits the broader community,” said Eric Boyd, Corporate VP of Microsoft AI and Research. “This is why Microsoft has collaborated with AWS to create the Gluon interface, and enable an open AI ecosystem where developers have freedom of choice. Machine-Learning has the ability to transform the way we work, interact, and communicate. To make this happen we need to put the right tools in the right hands, and the Gluon interface is a step in this direction.”

The pair say that the reference specification allows Gluon to work with any deep-learning engine, and that support for Apache MXNet is now available, with Microsoft’s Cognitive Toolkit (CNTK) soon to be added to that list – and we imagine Tensorflow won’t be far behind. Aimed at both cloud, edge, and mobile devices, developers should be able to use a ‘simple’ Python API and a range of pre-built neural network components, to swiftly build their applications.

Available through GitHub, AWS and Azure explain that developers build neural networks using three components – the training data, a model, and an algorithm. That algorithm trains the model to understand patterns in the training data – and the pair say that because the volume of training data is so large, and the algorithm is so complex, training the model takes weeks.

In addition, the pair note that while the deep-learning engines, like Apache MXNet, Microsoft’s CNTK, as well others including TensorFlow and Caffe2, have emerged to help optimize that process (for example, inferring whether the data needs to be re-computed, or if a tweak can be left as-is), they still require the developers to use complex code to define both model and algorithm – and that simplifying those two components has the trade-off of increasing the training time.

And lo, in step AWS and Microsoft with Gluon, providing what they say is the best of both worlds – “a concise, easy-to-understand programming interface that enables developers to quickly prototype and experiment with neural network models, and a training method that has minimal impact on the speed of the underlying engine.”

Additional benefits of the Gluon approach are that developers can tweak their neural networks ‘on the fly,’ as Gluon brings the training algorithm and the neural network’s model together, so that the developer can carry out the model’s training one step at a time. AWS and Microsoft claim this makes it much easier to debug, update, and reuse these neural networks.

The potential of machine-learning can only be realized if it is accessible to all developers. Today’s reality is that building and training machine-learning models requires a great deal of heavy lifting and specialized expertise,” said Swami Sivasubramanian, VP of Amazon AI. “We created the Gluon interface so building neural networks and training models can be as easy as building an app. We look forward to our collaboration with Microsoft on continuing to evolve the Gluon interface for developers interested in making machine-learning easier to use.”

As for supporters, the Financial Industry Regulatory Authority (FINRA), Carnegie Mellon’s School of Computer Science, the Georgia Institute of Technology, and Docomo Innovations, are all quoted in the release, singing the praises and promise of Gluon.

As for the name, a gluon is an elementary particle, which apparently acts as the exchange particle for the strong force between quarks. Theorized in 1962 and discovered in 1978, a gluon, as the name suggests, ‘glues’ quarks together, which allows them to form protons and neutrons.

Close