Many businesses these days are thinking about putting a toe into the waters of machine learning. They’re hearing a lot about this new kind of technology – and when you hear about brand-new technologies, you want to get in on the ground floor. You want to be an early adopter of the latest thing to stay ahead in the competition.
What a lot of companies will struggle with is that machine learning and artificial intelligence aren’t really user-friendly technologies. Unlike cloud computing, which typically involves just signing on for cloud-delivered services, machine learning requires some thought on the part of the buyer.
One of the big questions, and a common starting place for newcomers, is whether businesses should pursue supervised or unsupervised machine learning. These two fundamentally different types of machine learning both offer different benefits and drawbacks.
Supervised and Unsupervised Machine Learning
At its core, supervised machine learning directs the program in key ways where unsupervised learning typically does not.
In an interesting Quora response to the question of how these two machine learning models are different, you can see poster Debiprasad Ghosh, making this analogy – think about your kids playing and interacting with their world throughout the day. If they’re supervised, someone is helping to direct their actions, keeping them away from fire and other dangers, and trying to make their behavior socially acceptable. If they’re unsupervised, they’re making all of their own decisions.
This isn’t a perfect analogy, but it helps to start approaching how supervised and unsupervised machine learning differ.
The Importance of Labels
Let’s go to a bit more of a technical description of both supervised and unsupervised learning.
The key difference is that supervised learning uses labels to help direct the machine learning program. Labels are applied to different pieces of training data, so that the program can understand more about what has to work with.
In unsupervised learning, there are no labels. Machine learning programs are going “off-road” – all they have to work with is the inputs. They’re making their own conclusions.
One of the most popular and common explanations of this is the fruit basket model. In a supervised machine learning setup, you show the machine what a banana is, or a cluster of grapes. You give the machine pictures to classify. The machine will take new data sets and apply these classifications.
In unsupervised learning, all you can do is use the properties of the fruits to identify them (think long, yellow pieces and small, purple spheres in clusters). In many senses, you’re flying blind when it comes to the kind of data crunching that machine learning systems do.
Start with Supervision
In the overwhelming majority of cases, for any company that’s not specifically a data science firm or some kind of research facility, it makes much more sense to start with supervised learning. Putting in the time and applying the labels helps make your machine learning processes much easier to work with, especially if you’re using them for traditional enterprise intelligence like customer intelligence, product planning or anything else.
The flip side of this is that unsupervised learning works more robustly to catch missing patterns or fill in the blanks – but most retail and commercial projects don’t really need this level of detail. What they need is ease-of-use – they need a system that delivers the insights in a way that human handlers understand.
This is just to trigger thinking about exactly how to craft new technologies for your business. Let WebSubstance help you to plan for “web 3.0” – the kinds of state-of-the-art setups that are coming down the pike.