AWS Machine Learning service makes its debut

Two months after arch-rival Microsoft Azure rolled out machine learning services, AWS has answered. Amazon also delivered a raft of other much awaited cloud services.

SAN FRANCISCO -- Amazon Web Services now has a machine-learning service modeled on the one used by, a move IT pros welcome as a potential game-changer for developing modern apps.

The company also launched Lambda into generally availability with new features, as well as the Amazon Elastic File System and new desktop cloud services, but it was machine learning that wowed the crowd here at the AWS Summit this week.

"I have done some machine learning stuff in the past, but it's always been kind of a headache," said Todd Blanchard, co-founder of Rewind, a social network due to launch later this month, based in San Diego. "As a small company, we don't have the luxury of hiring a guy to just sit there and figure out machine learning."

The new service will allow for a more polished friend-recommendation service for Rewind, Blanchard said, as well as a potentially updated parking management system for another company he works with called iParq.

"The machine learning will have immediate usage from us," said Kevin Felichko, CTO of, an online auction company based in Frederick, Md., who considered Azure Machine Learning before Amazon introduced its own..

One of the biggest challenges for PropertyRoom is fraud, and machine learning could help separate legitimate from fraudulent users, Felichko said.

Amazon Machine Learning is a fully managed service aimed at developers with little experience in machine learning concepts. It automatically finds patterns in data and allows developers to do model building, validation and optimization, and to make predictions asynchronously in batches or synchronously through an API.

Data analysis, model training, and model evaluation will cost $0.42 per compute hour; batch predictions will cost $0.10 for every thousand predictions, rounded up to the nearest thousand; real time predictions cost $0.10 for every thousand predictions plus an hourly reserved capacity charge of $0.001 per hour for each 10 megabytes of memory provisioned for a given model. Charges for data stored in S3, Amazon RDS, and Amazon Redshift will also apply. It is generally available now.

AWS Lambda generally available 

Amazon Machine Learning was far from the only news here this week; AWS also updated Lambda, now generally available, as well as the Amazon Elastic File System, a new NFS-based interface for its storage systems, both of which were also welcomed.

"Lambda makes things that used to require a huge complex compute platform accessible to a wider array of developers," said Mark Buster, principal designer for Leviton Manufacturing Company Inc., a privately held manufacturer of electric wiring equipment based in Melville, N.Y.

Several of the new features in Lambda were on Buster’s wish list already, such as the ability to trigger functions from the Simple Notification Service (SNS), as well as a tie-in to the Amazon Cognito mobile development platform and support for Java commands as well as node.js. These updates will allow Leviton to use the service for a wider range of applications than the preview version.

"It makes things a lot easier – we won’t have to maintain dedicated EC2 instances, for example," Buster said.

The Amazon Elastic File System also drew enthusiastic applause from attendees at a keynote here. It can scale to petabytes of data and is automatically replicated across availability zones and is compatible with NFSv4. EC2 instances in the same region can access the file system regardless of their availability zone; the file system will support thousands of concurrent NFS connections. The preview for this service will start this summer.

With a warehouse in New York, offices and Maryland and remote workers scattered in between, PropertyRoom will look into the shared file system to replace local network attached storage (NAS) devices, Felichko said.

"File sharing on those devices is slow and cumbersome," he said.

Container Service, desktop apps also get a brush-up

In a product-update-packed keynote, AWS officials also said that the EC2 Container Service (ECS), announced at last November’s re:Invent conference, is also generally available with some updated features, such as the ability to mount containers on Elastic Block Store volumes, a new service scheduler for long-running applications, private Docker repositories and integration with the CloudTrail API audit service.

Felichko said he’d hoped for such a scheduler to be included with Lambda.

"I can see us deploying Lambda, especially as we move into the mobile space, but I wish there was a scheduler or [Amazon Simple Workflow Service] integration – that would make it much more convincing for me."

Amazon ECS is available in the Asia Pacific (Tokyo), U.S. East (Northern Virginia), U.S. West (Oregon), and Europe (Ireland) regions.

Finally, Amazon delivered services at the desktop level, namely a new marketplace for pay-by-the-month desktop applications and a WorkSpaces Application Manager that can package and manage virtual applications for Amazon WorkSpaces desktops, including automated patch, backup and upgrade management.

Beth Pariseau is senior news writer for SearchAWS. Write to her at [email protected] or follow @PariseauTT on Twitter.  


Dig Deeper on AWS machine learning