Special Report: Artificial intelligence apps come of age
A comprehensive collection of articles, videos and more, hand-picked by our editors
Over the past year, the cloud has become much friendlier terrain for machine learning applications.
Microsoft and Amazon Web Services both made cloud-based machine learning platforms generally available in early 2015, giving them very similar names -- Azure Machine Learning and Amazon Machine Learning, respectively. And later in the year, Databricks and IBM each launched a cloud version of the Spark processing engine, which includes a built-in machine learning library.
Cloud setups can be particularly useful for testing cloud machine learning algorithms on live data sets, since users don't have to provision and configure in-house systems for the trial runs. "It's a cheap way to test things out," said Joe Emison, founder and CTO at BuildFax Inc., an Asheville, N.C., company that collects data on building permits nationwide and provides analytics services on it for homebuyers, insurers, realtors and inspectors.
Emison said developing machine learning applications tends to involve a lot of trial and error, which makes it helpful to be able to test modeling concepts and designs in an inexpensive way. But BuildFax has gone beyond that stage in the cloud: It uses Amazon Machine Learning in production for predictive modeling against data stored in a MySQL database running in the AWS cloud as part of the Amazon Relational Database Service. The company's models can predict, for example, the age of roofs on houses in certain areas. How old a roof is, Emison noted, can sometimes be an effective indicator of the overall risk of insuring a home.
For midsize companies like BuildFax, Emison thinks opting for turnkey cloud services is preferable to building on-premises machine learning systems themselves. But there's no free lunch, he said, adding that cloud or no cloud, analytics teams still have a lot to learn about how to make machine learning work. According to Emison, that includes boning up on statistical analysis methodologies and what makes an effective algorithm.
At commercial insurance company Zurich North America, running machine learning applications in the cloud is still a topic of internal dialogue. "We do everything on premises now," said Conor Jensen, the Schaumburg, Ill., company's analytics program director. "But it's an ongoing discussion."
Machine learning democratizes big data
Big data ecosystem breathes life into machine learning
Machine learning should be available to all
Making sense of big data with deep learning
An expert take on the advances in machine learning platforms