Microsoft Azure Machine Learning (ML) is a rich environment for building data-driven models that “learn” how to solve difficult problems such as those found in data classification and time-series prediction. There is currently a surge of interest in machine learning, because new algorithms, in particular deep learning, have created a step improvement in accuracy over earlier methods, leading to applications in areas such as image recognition, speech processing, and others. The key challenge for a service provider is how to make these methods easy to use without needing a PhD in ML. Ovum’s first impression is that Microsoft has successfully overcome this hurdle. Certainly, the customer use cases that were recently showcased at Microsoft Build 2015 in San Francisco demonstrated how ML-infused applications can create new businesses and breakthroughs.
Azure ML has three layers for graded ease of use
The Azure service offers three levels for different types of user. The easiest to use layer enables data scientists to pick existing solutions from a gallery of pre-built solutions on gallery.azureml.net, and easily tweak these in Azure ML Studio, which can be thought of as a Visual Studio-like environment for building ML models, with visual drag-and-drop features. At the second layer, developers can dive deeper in ML Studio and open up a menu of algorithms to create custom solutions. At the most sophisticated level, experts in ML can add their own code and combine these with the pre-built ML library, in ML Studio, or via its API.
ML is on the brink of infusing intelligence in mainstream applications
While ML has been used in niche areas such as fraud detection in banking for some years, the mainstream use of ML in everyday applications has not yet taken place, but this is likely to change. A number of cloud providers now offer ML services for integration in applications, and Azure has a comprehensive set of algorithms available and wide range of types of models that can be used, from completely pre-built applications to models that can be fully customized. A popular Microsoft example is how-old.net where users can upload a photo of themselves and it will tell them their age based on the eyes-to-mouth area. (In my case, it made me nine years younger, but this model made no use of hair, or lack of it, information, so it could be improved).
Also showcased at Microsoft Build 2015 was eSmart Systems, a Norwegian startup that has built its solutions on Azure using ML. The company’s Connected Grid solution helps electricity grid generation companies manage their services more efficiently.
Ovum Analyst Insight, Machine Learning in Business Use Cases, IT0022-000335 (April 2015)
Ovum Opinion, Open source is accelerating artificial intelligence innovation, IT0022-000328 (March 2015)
Michael Azoff, Principal Analyst, IT Infrastructure Solutions