Amazon has launched a public cloud version of the same machine learning service it uses for its own retail store. Amazon Machine Learning requires no specialized skills on the part of the user, and allows businesses to quickly use historical data to build and deploy predictive models, the company says.
The models can be used for a variety of purposes, including detecting problematic transactions, preventing customer churn and improving customer support. Amazon Machine Learning is based on Amazon's own internal tools, generating more than 50 billion predictions weekly, Amazon says. (See Amazon Web Services Debuts Amazon Machine Learning.)
The new service integrates with Amazon Simple Storage Service (S3), Amazon Redshift and Amazon Relational Database Service to allow customers to use data they already store in the AWS Cloud, Amazon says.
Machine learning has previously been difficult, requiring specialized expertise, Andy Jassy, senior vice president, Amazon Web Services Inc. , said in an announcement Thursday that was streamed over the Internet (with some facepalm-inducing glitches -- see D'oh! Amazon Cloud Fails at Its Cloud Summit).
"It's hard work and it requires machine learning experts internally," Jassy said.
Amazon's own Amazon Machine Learning automates the process, requiring no specialized skills from developers.
Amazon uses the same engine in its own customer recommendations, as well as item classification, counterfeit goods detection, sales lead rankings, determining search intent, estimating demand, providing customer support, displaying ads, voice recognition and to unload trucks of products quickly and make them available for purchase, Amazon says.
As an experiment, Amazon pitted conventional machine learning tools against its own service. The company assigned two developers to learn machine learning, and build a tool that would determine the gender of customers. Some names are unisex, like "Pat" or "Jerry," Jassy explained, making the problem complicated. Using conventional machine learning software, the engineers were able to build a tool in 45 days that determined the gender of customers with 92% accuracy.
Using Amazon's Machine Learning service, a single developer, also inexperienced with Machine Learning, was able to write software to solve the same problem with the same degree of accuracy -- 92% -- in an hour, Jassy said.
"Think about how enabling that is, to start to do machine learning across your entire development team without having to be a machine learning expert and do it with a high degree of accuracy," Jassy said.
Amazon cites Comcast Corp. (Nasdaq: CMCSA, CMCSK) as a reference customer for Machine Learning, and says the cable provider is using the service with Comcast Cable and NBCUniversal.
Amazon is not the first to offer machine learning as a cloud -- Microsoft launched Azure Machine Learning last year.
Streamlining virtual desktops Amazon has also introduced AWS Marketplace for Desktop Apps, and Amazon WorkSpaces Application Manager.
The two services are designed to ease the burden for IT of managing desktop PCs. Enterprises moved to virtual desktops to reduce the cost and hassle of managing desktops remotely, but found that even centralizing the desktops in the cloud was still expensive and a lot of work. Amazon's own VDI service, WorkSpaces, is designed to reduce that cost, Jassy said.
The Marketplace for Desktop Apps supports more than 100 applications in 11 categories, such as Security, Productivity and Collaboration, Business Intelligence, and Illustration and Design. The Marketplace allows users to pay by the month for easy cost management, rather than requiring annual or perpetual licenses. The Marketplace doesn't require additional contracts or invoices and customers receive all software charges in their monthly AWS bill, Amazon says.
WorkSpaces Application Manager packages and delivers apps to WorkSpaces so they run as if natively installed but are centrally controlled by IT administrators for simplified maintenance and auditing.
Many companies deploy more than 200 desktop applications, requiring infrastructure to manage the software and track usage, as well as managing upgrades and retiring apps when they're no longer needed. Amazon's WorkSpaces Application Manager is designed to simplify that process, Amazon says.
Why this matters The Machine Learning tool will be useful to carriers as well as other enterprises. Carriers face the same problems as other businesses, designed to be addressed by machine learning, such as customer churn, fraud, and recommendations.
Moreover, both the machine learning and desktop products will make AWS more attractive. As carriers build businesses of connecting enterprises to cloud providers like AWS, anything that makes clouds more attractive helps carrier business.
More on cloud: