How to approach machine learning in the cloud

Machine learning needs lots of data, and the best place for all that data and the systems that use it is in the cloud

How to approach machine learning in the cloud
Credit: Pexels

Artificial intelligence and its machine learning subset are all the rage these days. That was evident when I spoke this week at the AI World event, which was packed with vendors and users seeking to understand what the hell AI and machine learning are—and wanting to know how they could use this old but revitalized technology effectively.

Amazon Web Services, Google, IBM, Microsoft, and the other major cloud providers all have machine learning services in their clouds now. But most enterprises have no clue on what the heck to do with machine learning systems, whether cloud or on-premises. Here’s some quick guidance.

The right use cases for machine learning get the real value

It is critical to find the right uses for machine learning. For the most part, simple applications that update databases and perform simple calculations are not a good fit. Just because machine learning is cheap in the cloud does not mean it should be applied everywhere.

Instead, you should seek repetitive tasks typically done by a person now. When such tasks are truly rote, you can automate them such as through a robot in an assembly line.

But many tasks people perform require judgement and analysis to determine the right solution. These tasks typically have patterns of discovery and resolution, so in most cases a person repeats the same set of diagnostics, then applies a known solution.

Machine learning can do that, too, automating the discovery and solution after using a knowledge engine to determine the patterns a human expert uses. A knowledge engine can learn as it operates, exactly like people, so the system can in some ways reprogram itself to get better at its job over time. And it’s cheaper than a person. 

Examples of tasks that are good fits for machine learning include fraud detection, risk analytics, maintenance scheduling, quality assurance, and medical diagnoses. (When an AI system can’t figure out the problem or solution, it then kicks the issue to a human expert.)

Bind machine learning systems to the data—in the cloud

Even when a task is a fit for machine learning, how do you know that automation should be done in the cloud? The most compelling reason is that you can colocate in the cloud your machine learning applications with cheap data storage.

Machine learning has no value without data. Data provides patterns to the knowledge engine, in the same way that a person may look through several reference books for patterns, or models, to better understand new and changed data. Thus, you need to bind machine learning systems to as much data as you can. The cloud is the best place to do that.

To comment on this article and other InfoWorld content, visit InfoWorld's LinkedIn page, Facebook page and Twitter stream.
From CIO: 8 Free Online Courses to Grow Your Tech Skills
Notice to our Readers
We're now using social media to take your comments and feedback. Learn more about this here.