Do data centers dream of electric sheep? Microsoft announces machine learning cloud
New Azure service sees Redmond open up a bit of its brain to developers
Microsoft is flinging some of its internal machine learning tech up into the cloud, putting its Azure service ahead of rival products from Google and Amazon.
The company's new "Azure ML" service was announced on Monday and means developers can access machine learning systems hosted in the Azure cloud and even link their applications directly to them.
The tech gives developers a directory of machine learning and associated technologies, including deep learning systems, that they can apply to their applications.
Recently, Google announced that one of its engineers had built a system that used a neural network to calibrate the relationship between 19 different inputs and a single output – data center power usage effectiveness. The system worked so well that Google was able to use it to better simulate how its power consumption would change over time.
With Azure ML, Microsoft says it is making these types of systems available to a broad swathe of developers by making it easier to set up, train, and use the tech.
Azure ML also has "a number of tools to help clean data," explained Microsoft exec Joseph Sirosh in a chat with El Reg, and has compatibility with popular mathematical software R. The service also gives users a way to drag-and-drop various machine learning technologies together so that they can build an application in a visually striking and understandable way.
"The whole universe of machine learning is open to you here," Sirosh said. "Within a small number of clicks you can get an API with machine learning models."
Sirosh came to Microsoft after spending almost a decade at cloud king Amazon, where he helped build some of the company's internal machine learning systems.
Some early customers of the service, which is expected to launch in Beta in a month, have been using it to predict what product online shoppers are likely to buy based on their actions on a site, and another is building a system that can work out how electrical fluctuations across a university campus can help it diagnose equipment faults.
Though machine learning is an old and widely used technology, Microsoft thinks that presenting it in a cloud-service gets rid of some of the barriers adopters of the tech have faced.
"Machine learning is an incredibly underutilized capability – every app around us could be becoming intelligent," Sirosh said. "I would love to have the excitement around machine learning be unleashed. It's just like the birth of the cloud."
And he thinks, naturally, that Microsoft is one of the best companies to be serving up such a service.
"Microsoft and Google are the repository of machine learning talent in the world today," Sirosh said. And for now, Google hasn't served up some of its fast ML systems on its cloud. ®