In this series of blogs, we’ve explored topics like AI and machine learning, and investigated their impact on IT operations. In today’s discussion, I’ll be offering a primer on two different versions of machine learning: supervised and unsupervised learning. Understanding these distinctions illustrates how far usable machine learning has come. And I’ll outline how unsupervised learning can make a big impact on your IT operations in at least three fundamentally important areas.
The defining characteristic of supervised learning is that both inputs and expected outputs are known in advance. You program a machine to recognize an input and train it to deliver the desired output. I sometimes use the example of colored flags to illustrate this point. You teach the machine to recognize blue flags, red flags, orange flags and so on, tuning the results again and again until you get 100% accuracy. This kind of process works fine with simple stimuli and basic tasks. But as we all know, real-world business and IT environments are anything but simple. For example, replacing the colored flags with colored balls would require one to start the supervised learning process all over again.
Instead of machines being hand-trained by data scientists, unsupervised learning uses algorithms that identify consistent, coherent and recurrent patterns in data. Once the algorithm identifies these patterns, it’s able to autonomously identify causality – i.e. relationships within data that flag when future issues are likely to occur. This is a key evolutionary step toward the future.
Machine learning built for the real world
Understandably, some IT leaders are skeptical about unsupervised learning, and whether it is truly capable of delivering meaningful insights about vast and complex enterprise IT infrastructures. It is. There are approaches that are designed to work at a massive scale for machine learning, and make unsupervised learning realistic for enterprise IT. These huge-scale, near real-time machine learning operations far exceed human capabilities, delivering the volume of personalized insights that only automation can deliver.
In fact, if I were identifying the top three advantages of unsupervised learning for IT operations, scale would be first. Second, the power of real-time analysis. For example, the ability to identify behavior and move workloads around dynamically in response to changing patterns in your infrastructure. Or, in a world where 3-day old data is too old, creating marketing offers in real-time based on credit analytics and recent shopping patterns. The third (and perhaps most important) advantage of unsupervised learning is how it applies to the real world of business. Here every enterprise boasts its own unique selling proposition, sales and marketing strategy and so on, and builds an IT infrastructure to deliver its business goals.
The bottom line? Even in the most complex, data-intensive environments, unsupervised learning will deliver actionable insights that are unachievable by any other means.