Machine learning is nothing new, in fact it’s nearly as old as the pursuit of artificial intelligence itself, but despite this it’s never really become what you’d call a mainstream application. It’s a well-worn story for many technological innovations that have existed as working concepts long before they became a part of modern life. The most advanced ideas are often ahead of their time and it has taken decades for computer power and networking infrastructure to reach a point where machine learning’s potential could be unlocked. The other enabling factor is of course the flood of data that’s now being generated from all fronts as we move towards the Internet of Things. Everyday more and more daily transactions are becoming digital and therefore these powerful machine learning algorithms have data to chew on, providing us with insights that were simply impossible to get before using more conventional techniques.
Unfortunately these powerful advantages have been reserved for those with the financial and technical resources the technology needs. Apart from this the actual implementation of machine learning, especially the custom creation of effective algorithms, can be a messy progress without a clear path from point A to point B. This is why machine learning applications that are of commercial value have largely been limited to organizations with large and flexible R&D budgets such as Google and Microsoft.
Microsoft Azure ML (machine learning) is set to change this in a big way. For years the trend towards software as a service (SaaS) and greater reliance on outsourced, cloud-based data centres have laid the groundwork necessary to Azure’s approach to ML. The way Microsoft has designed Azure ML is essentially “machine learning as a service”. There are pre-existing algorithms and supercomputing power just a few clicks away while you can still use custom R code should you need to. In general however, the service is designed to minimize or completely eliminate coding for the vast majority of user’s. Doing for the obscure world of predictive analytics what the GUI did for DOS.
From a business perspective the cost/benefit argument for or against ML has been turned on its head. Where each individual enterprise was left to essentially re-invent the wheel, services such as Azure ML allow for rapid deployment without any need to procure capital, train staff or purchase software licenses. In other words the service is “fully managed” according to Microsoft, which also implies that various forms of support are also included as part of the value proposition. The most valuable part of the whole offering might very well be the immense speed with which solutions can be put into practice. Successful models in the virtual experiments area of Azure ML can be automatically turned into a working web service within seconds.
There can be little doubt that Azure ML is likely a sign of things to come. Expect competitors to enter the market sooner rather than later. The time is ripe to start asking questions about how you would implement machine learning and predictive analytics in your own enterprise or business, given the chance. You can be rest assured that your competitors are probably doing the same.