Machine learning as a service isn't new, but when Microsoft offers it as part of Azure you have to take notice.
At the moment Azure ML is only a preview for Microsoft partners, but in July a public beta will be launched and you can sign up now to gain access to it.
It is also worth noting that Microsoft already has a cloud based service for big data based on Hadoop - Azure HDInsight. Azure ML seems to be targeted at users not willing or able to learn the technicalities of implementing a Hadoop solution.
What isn't clear at the moment is what techniques will be included. You can probably make a list of the obvious candidates from regression to Bayes decision trees and deep neural networks but the following quote from the Microsoft blog makes it slightly more interesting:
"Azure ML, which previews next month, will bring together the capabilities of new analytics tools, powerful algorithms developed for Microsoft products like Xbox and Bing, and years of machine learning experience into one simple and easy-to-use cloud service. "
What new analytics tools and algorithms could they be referring to?
The front end looks easy to use and is called, predictably, ML Studio. It seems to have a graphics designer and it refers to each analysis as "an experiment".
You can see what Microsoft has in mind from the following, very over the top, promo video:
The first thing you notice is that the emphasis is on "predictive analytics" which is something we used to call "statistics", but I guess that would make it sound very boring. According to Microsoft the power of predictive analytics is no less than miraculous and enables you to make the railroads run on time, or at least the modern day equivalent the plane, and detect a fraud before it happens. It seems that Microsoft partners have been using Azure ML for some of these tasks
"Today, partners are using an early preview of Azure ML to build machine learning solutions for our customers. For example, MAX451 is helping a large retail customer determine what products a customer is most likely to purchase next, based on ecommerce data as well as brick and mortar store data. OSISoft is working with Carnegie Mellon University on real time fault detection and the diagnosis of energy output variations across campus buildings. Machine learning is helping to mitigate issues in real time and to predictively optimize energy usage and cost."
It all sounds too good to be true and it probably is. I wonder how many innocents are going to try to extract information from their data only to jump to the wrong conclusions because of inadequate understanding of basic statistics? There are many things that ML is good for, but it isn't a replacement for basic stats.
Google has finally made a move on its promise of Android for Work, first announced at last year's I/O conference. It has a range of features that make Android more business-friendly, but there are som [ ... ]