Machine learning is based on old artificial intelligence concepts. It was first defined in 1959 as the ability...
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
to give systems the capacity to learn without constantly having to update them. It rose out of pattern recognition and computational learning, and has recently stepped into the spotlight with major public cloud providers now offering their own machine learning services.
Today we know machine learning as the study of algorithms that have the capability to learn by seeing patterns, typically in data. Many view it as a superior approach to static programming. It simulates the way humans learn through experiences, improving over time without the manual addition of new code or other configuration information.
The cloud has made machine learning services more available, affordable and useful. We now have access to compute and storage that we rent, not buy. You can use petabytes of data for machine learning applications at a fraction of the cost of on-premises hardware. The renewed interest in machine learning stems more from the new capabilities and costs of the technology than the old capabilities of machine learning itself.
Common machine learning use cases include fraud detection, inventory management and even the ability to control machines within internet of things applications -- pretty much anything that could benefit from the knowledge of data patterns.
There are other, more sophisticated applications of machine learning, as well. For example, in the MIT Sloan Management Review article Sales Gets a Machine-Learning Makeover, the Accenture Institute for High Performance shared the results of a survey in which enterprises with greater than $500 million in sales participated. Survey respondents were asked if they planned to generate higher sales growth with the technology and 76% said yes. To do this, they plan to use machine learning for greater predictive accuracy, and to align sales resources accordingly.
Machine learning services in the public cloud
Google Cloud Machine Learning on the Google Cloud Platform and Amazon Machine Learning on Amazon Web Services (AWS) are the leading examples of public cloud machine learning services. Both offerings apply machine learning technology within their respective environments to drive interest in application development on their cloud platforms. Both provide the ability to use machine learning services, as well as the big data management systems that serve as the data source, at a low cost.
When choosing a provider, consider all aspects of your machine learning requirements and how the public cloud provider can meet them. Beyond the actual machine learning services, consider the way in which data, middleware and analytics on the cloud platform will work together to solve business problems.
Machine learning systems from public cloud providers offer software developers' kits and application program interfaces that allow developers to embed machine learning capabilities within applications. This bridges the gap between the capabilities of machine learning and the actual use of the technology. For example, an organization could determine if a loan application is fraudulent based upon past and current data patterns within the loan application.
There are downsides to machine learning services in the public cloud. First, it requires the use of a service that is native to the public cloud provider, meaning you have to port the data to other clouds or back on-premises, which could be problematic.
Break down new machine learning methods
Jack Vaughan and Ed Burns discuss machine learning methods that are poised to change the course of data analytics.
Second, many enterprises have a tendency to overuse machine learning, implementing it for applications that don't actually require the capabilities; machine learning is overkill for simple business processes that are more procedural in nature. Applications that perform simple structured and sequenced processing are typically not a fit for machine learning. For example, applications that book sales, track shipments and handle other well-defined processes won't really benefit from a learning engine.
Of course, you have to consider all application requirements before making such an assessment. Before jumping in, carefully consider the different machine learning services from providers like AWS and Google, and remember that not all applications are a good fit for the technology.
Machine learning and IoT dominate cloud market
Learn the benefits of a machine learning service
Compare public cloud providers' big data services