There is more than one way to build a piece of furniture. The end result for each should be the same, a completed desk or bookcase for example. How you choose to get there, though, may not be straightforward.
If you have the instruction manual it is as simple as following the steps. Maybe you have built one before and decide to go from memory. Or maybe it never came with a manual and you have to figure it out through trial and error.
Machine learning works in a similar fashion. The end result you are going for is a well-trained algorithm, but there are different ways to get there. Two of the most common through supervised and unsupervised learning.
Supervised Learning
Much like it sounds, supervised learning means something is watching over the results to determine whether they are correct or not. In machine learning, this means that the algorithm needs a complete set of labeled training data to work with.
Labeled training data means that each correct answer, the outcome you want the algorithm to come to, is tagged accordingly. For example, if you were trying to train an algorithm to correctly recognize photos of sofas, the training data would tell it which supplied pictures were sofas rather than chairs or bookcases. The algorithm then compares every new image to what it has already learned about the original images and makes a prediction.
Supervised learning is most commonly used for regression and classification.
In classification, the algorithm tries to analyze the new data it is fed, then assign it to a certain group or class. Sticking with the furniture analogy, the labeled training data would tell the algorithm which pictures were of sofas, which were chairs, and which were bookcases. The algorithm then takes new images, attempts to classify them into one of those groups, and is scored based on how accurately it did so.
Regression, on the other hand, deals with continuous data. It is more useful when you have multiple data points you are trying to train the algorithm on. Maybe you don’t have photos of the different types of furniture, but information about the furniture instead. You could feed data such as the cost of the item, the width, height, length, and weight, and train the algorithm to make the prediction based on that.
Supervised learning is great when you have reference data to train with. When you don’t, you may have to turn to unsupervised learning.
Looking for career tips from Chicago’s top IT staffing team?
What is a Hiring Surge?
5 Skills Needed to be Successful in Today’s Workforce
The 4 Biggest Mistakes to Avoid When Preparing for an Interview
Unsupervised Learning
When you don’t have fully labeled training data, or are asking questions you don’t know the answer to, you’ll likely have to turn to unsupervised learning.
In unsupervised learning, the algorithm is given data without much information about what to do with it. In other words, there is no known “right” answer. It is up to the algorithm to decide what to do with it. It tries to analyze the data looking for patterns, structure, commonalities, anomalies, and features. It will then try to display what data points it thinks belong together or are connected in some way, and which are not. There are quite a few ways it can do this:
- Clustering The algorithm tries to find data points that are similar and cluster th