Microsoft CEO Satya Nadella keeps saying that Microsoft's Azure cloud platform makes it easier for firms to exploit machine learning (ML). But how far is this marketing message borne out by the services available on Azure?
Azure's suite of machine-learning offerings is fairly comprehensive, targeting everything from companies seeking simple, on-demand services through to those looking to train their own models using in-house data scientists.
Every platform-as-a-service (PaaS) machine learning-related product and service that Microsoft offers is part of the Cortana Intelligence Suite. This bundles Microsoft's analytics and ML-focused offerings with Microsoft cloud-based data stores, capable of holding the vast amount of data needed to train machine learning models. Other services transform data to prepare it for analysis, while others help to display the results clearly.
Azure's suite of machine learning services is easier for businesses to get grips with than offerings on some competing cloud platforms says Peter Chapman, head of capability at Rolls Royce, which makes use of various Azure machine learning services.
"Microsoft is more targeted at enterprise and platform-as-a-service, which suits our goals better," he said, stressing the suite's simplicity compared with Amazon Web Service's (AWS) more DIY infrastructure-focused offerings.
For those businesses wanting to roll their own analytics and machine-learning platform, the Cortana Intelligence Suite includes HDInsight, a managed version of the Hadoop big data platform, which includes a variety of software for processing and storing data at scale, as well as allowing automated provisioning of clusters via PowerShell. Data can be held externally using services such as Azure Data Lake Store, which offers high-performance storage based on the Hadoop Distributed File System (HDFS) standard. Pricing is linked to the number and type of nodes.
Firms that want less responsibility for managing the underlying infrastructure can use Azure Data Lake Analytics, which is designed to run analytical jobs at very large scale -- over exabytes of data if necessary -- and automatically takes care of the management and provision of underlying clusters. Data Lake Analytics is optimized to process data in Azure's Data Lake Store, but can also process data held in Azure's Blob Storage, SQL Database and SQL Data Warehouse. Pricing is based on how many analytical units are used and the number of jobs completed.
Real-time data processing can be carried out using Azure Stream Analytics, which can query data as it's collected using an SQL-like language or feed it into machine learning models for analysis.
Routing data where it needs to go is a task for Azure Data Factory: this service builds data-processing pipelines that can pipe data through services, transforming it in transit and interacting with other Azure services -- triggering the creation of HDInsight clusters on demand, for example.
Helping simplify the process for firms building their own machine learning model is Azure Machine Learning. Launched in 2014, this managed data science platform provides many of the components needed to build machine learning models. The bundled Azure Machine Learning Studio offers a graphical tool that lets users link together pre-built modules that handle each stage of training for machine-learning models. Users can drag and drop modules that link to datasets, pre-process data, run machine-learning algorithms and refine machine learning models. The platform allows users to build their own modules written in R or Python code. Once trained, models can be published as web services, which can be linked to by applications and respond to requests on-demand or during batch processing.
Firms requiring in-house data scientists to write their own analytics scripts using the statistical programming language R can use Microsoft DeployR to more easily integrate these models into applications. DeployR provides a framework for exposing these R analytics models as web services.
Making the output of these machine learning and other analytics models easier to digest is Power BI, which can take data from various Azure, third-party services and on-premise sources and present them in dashboards and other visualizations, including user-created offerings.
Organizations that don't want to do any of the work training up machine learning models themselves can use Azure Cognitive Services, which offers a suite of on-demand web services offering speech, vision, natural language and knowledge processing that can be built into applications and bots. Most of these services are billed per 1,000 API calls.
Finally the Microsoft Bot Framework packages up services for building a chatbot, including the relevant speech and language recognition APIs from Cognitive Services, plus services for hosting and deploying bots and for communicating via messaging platforms such as Skype, WhatsApp and Facebook Messenger.
Microsoft has built up the expertise that underpins these services over the course of many years, according to Herain Oberoi, senior director for product management for Cloud and Enterprise at the software giant.
"Machine learning is core to Microsoft's DNA and it has been for a long time, out of necessity because of what we had to do with our consumer businesses," said Oberoi, referencing the research that went into developing the Bing search engine, recommendations in Microsoft's Xbox games console, the visual recognition systems in the Xbox Kinect and translation engines in the communication platform Skype.
That level of investment continues to build, and in September this year Microsoft announced the formation of the AI and Research Group, led by Harry Shum, a longstanding AI researcher at Microsoft who helped developed the Bing search engine.
The cloud-based machine-learning marketplace is increasingly crowded, with competing services from the likes of Google, IBM and Hewlett Packard Enterprise. Mike Gualtieri, VP at analyst house Forrester, said that while Microsoft offered simpler tools for firms building their own machine-learning models, the quality of the firm's on-demand, pre-trained speech, vision and language recognition services would likely be less effective than Google's because of the search giant's access to huge amounts of training data.
For Rolls Royce's Chapman, part of the appeal of the Azure-based offering is that the cloud platform allows data to be kept in any one of 30 different regions where Azure is available worldwide: "Microsoft's regional strategy for cloud works well for our business needs, for data protection and data privacy, being able to have clear country boundaries," he said.
Rolls Royce uses the Cortana Intelligence Suite's HDInsights capabilities to aggregate data from a large number of sources, as well as the Azure Machine Learning platform and the Data Factory processing pipeline, which together power services that predict fuel consumption and maintenance requirements of its airplane engines.
The firm is one of many customers from a wide range of industries that use Cortana Intelligence Suite services: including German elevator multi-national ThyssenKrupp, online transport network Uber, Carnegie Mellon University, soccer giant Real Madrid and American football's NFL.
As with any service, there are drawbacks to Cortana Intelligence Suite -- in particular, the fast pace at which new services come online on the Azure cloud platform.
"This is not unique to Microsoft. Clearly things are maturing fast, so there's a rapid change of technology and what we've found is that some of the services are maturing at different rates to each other," said Rolls Royce's Chapman.
"While they've each got their own individual roadmaps, they might not have their roadmaps completely aligned to our use case, and our job is to try to knit these things together. We're often facing decisions on should we use this technology now, when in two months time something else is coming along that perhaps is going to be better suited," he added.
Ultimately, Forrester's Gualtieri predicts that Azure-based machine learning services will be continue to be a central pillar of Microsoft's ongoing reinvention as a cloud-first company.
"Microsoft understands that, in future, applications are going to have this intelligence and so strategically they have to be able to play there. This is not a casual investment, it's a hardcore investment."