Since the first computers blinked and beeped their way into science fiction movies, machine learning has been viewed...
as the future of computing. Today, that future is here, as machine learning -- especially in the cloud -- becomes increasingly practical and affordable for enterprise IT.
More affordable machine learning services in the cloud, particularly from leading vendors like Amazon Web Services (AWS), Google and Azure, have democratized the technology, making it available to more organizations and for more purposes than ever before.
Any machine learning system needs three main ingredients: lots of inexpensive compute power, lots of data for analysis and a run-time framework to coordinate algorithms, said Greg Arnette, founder and CTO of Sonian, a company in Waltham, Mass. that partners with AWS, and also uses AWS machine learning to deliver Sonian's cloud archiving and analytics services to customers.
The services available in the cloud today provide a broad set of these capabilities, as well as other machine learning tools that were previously only available to large corporations, universities and governments, Arnette said.
Machine learning services in the cloud vs. DIY
The benefits of pursuing any machine learning use case -- including image and pattern recognition, document analysis, forensic analysis and digital loss prevention analysis -- in the public cloud are primarily low cost, high reliability and fast performance.
"Trying to replicate a machine learning processing framework on premises or in one's own data center will cost a lot and not perform as smoothly," Arnette said.
Ultimately, the selection of cloud or on premises "is no different from any other DIY choice," said Dan Conde, an analyst at Enterprise Strategy Group in Milford, Mass. What the cloud providers offer may not be perfect, but users "can just add a few lines of code to further exploit it," as opposed to having to weave it from scratch on your own premises, he added.
AWS vs. Azure vs. Google
While most cloud providers with machine learning services offer pay-as-you-go options, algorithms, APIs, data and modeling tools and fully managed infrastructure, it is still critical to choose the right provider, said Charlie Li, chief cloud officer at Capgemini, a consulting firm headquartered in Paris. That decision, however, can sometimes be confusing due to different pricing structures, security features and other factors.
"If your team is new to machine learning, look for a vendor who can provide out-of-the-box [options] and not those with features more suited to advanced users," Li said. Also, determine how easy it is to manage your data with a certain provider. New machine learning users, in particular, should ensure they can import data quickly and easily. They should also determine whether they can manage the data directly on the cloud platform or if they'll require outside tools.
Charlie Lichief cloud officer at Capgemini
Generally, Google, Azure and AWS machine learning services are more alike than different. Each offers a range of services to help with modeling and orchestration, as well as databases and storage to facilitate rapid analysis, Arnette said.
Some further distinctions among the services, according to Arnette, are:
- Google Compute Platform uses the popular TensorFlow framework to appeal to developers that might otherwise go to AWS or Azure, he said. TensorFlow is a platform that orchestrates machine learning jobs, allowing researches to spend more time on their algorithms and less time on the undifferentiated plumbing. For this reason, "[Google] appeals to the hard-core data scientist," he said.
- Azure has gained credibility and is especially popular with enterprises and .NET programming shops.
- AWS has many options for elastic compute and databases as a service. Like Azure and Google, AWS also has GPU support for computational-intensive workloads. The vendor also offers Rekognition for image analysis and APIs. In general, "AWS [machine learning] appeals to startups and non-Microsoft-centric teams."
Michael Fauscette, chief research officer at G2 Crowd, a business software review platform based in Chicago, said Azure seems to be the most feature-rich for building machine learning and artificial intelligence (AI) environments from the ground up, offering robust development tools and a library of prebuilt components. Google and AWS, however, have a more robust set of prebuilt components at this point, since both opened up APIs to existing AI functions that they had built for their own use.
For this reason, Google and AWS tend to be the most accessible for teams that are less technically inclined, while Azure provides deeper and more customized machine learning services.
"[Azure] not only has a marketplace for vendor offerings within its ecosystem, [but] also has a place for users to share code," Fauscette said.
Looking at the evolution of machine learning and its effect on the enterprise, Fauscette sees two main takeaways. The first is that, according to surveys his company conducted, there's a reduced level of concern among hiring executives about the shortage of data scientists. "My theory is that this is connected to the growing availability of easier-to-use machine learning capabilities," he said. What's more, as machine learning services become more popular, they open a new frontier for artificial intelligence.
"We are moving from analytical looks at historical information to predictive analytics and now to proscriptive support, where the machine communicates with the human about the optimal decision to make," Fauscette said. That capability is on the horizon in the cloud, as well as through traditional enterprise vendors like SAP and Oracle, he added.
Evaluate big data services from top cloud providers
Find out what Google offers for machine learning and big data
Create machine learning apps for AWS