If you’ve read my free e-book ML Simplified (tsk tsk if you haven’t), or read this post, you should have an idea of what Supervised Learning is.
But let’s peel the wrapper back a bit more. Actually, a lot more.
Supervised Learning basically gives your computer a cheat sheet during training. You’re telling it: “Here the question AND the answer buddy, go study. Don’t f*** this up”.
Its one of the most essential parts of ML. Its like the machine learning equivalent of open-book tests.
Allow me to supervise your brain model, and let’s dive in.
📗 What Is Supervised Learning?
Supervised Learning is often the first stop for those entering the world of ML.
Why?
Because it offers a straightforward approach that gives you the confidence boost before going into something more complex (ahem..unsupervised learning).
Its is essentially the teacher-student dynamic in the ML universe. Its like giving your model a neat set of flashcards where each card has a problem on the front and the solution on the back.
The models learn from each card, pairing each input (problem) with a pre-determined output (solution).
Eventually when you give the model enough “flash cards”, it turns it from this:
Wholesome Newbie.
To this…
Making Progress…
To hopefully this:
Supervised Enlightenment.
Here’s how Wikipedia defines it:
Supervised learning (SL) is a paradigm where a model is trained using input objects (e.g. a vector of predictor variables) and desired output values (also known as a supervisory signal), which are often human-made labels. The training process builds a function that maps new data to expected output values.
Supervised Learning is smart, adaptable, and great at making sense of relationships in data — kind of like a super-powered matchmaker for data points.
For example, you can give the algorithm a bunch of photos of popular fruits with labels (“apple”, “banana”, “orange”, “Ricky Martin”, etc.), and it learns to recognize them from their features.
Later when you show it a new photo, it tries to guess the label based on what it learned.
In short: Supervised Learning = Input + Correct Output → Learn The Pattern.
📚 The Core Components
Supervised Learning has a few core components that make the magic happen, that guide algorithms through examples until they’re ready to make predictions independently. Let’s look into each of these components.
Labeled Data
Definition: Labeled data consists of input data paired with the correct output.
Imagine a museum where each art piece has a tag with its name and artist. Similarly, labeled data comes with clear identifiers, showing the correct answers for training. These labels help the model understand what it’s seeing.
Take a look below for instance, with all the shapes with their assigned labels:
Image Credit - Karyna Naminas
Training
Definition: Training is the process where the model learns from labeled data to discern patterns.
Think of training like rehearsals before a big play. The model repeatedly processes the labeled data, learning to map inputs to outputs accurately. Each repetition polishes its ability to predict outcomes, establishing a mental framework to interpret new data effectively.
Testing
Definition: Testing evaluates the model’s accuracy using new, unseen data.
Picture testing as the models’ final exam. Here, its challenged with data it hasn’t encountered before to see how well it applies what it learned. The goal is to access the model’s ability to generalize from training to real-world data, ensuring it’s ready for practical applications.
🔑 Key Applications
Here are a few examples of where supervised learning is used:
Spam Detection
Annoying emails beware; supervised learning models will eject spam like they’ve been pro bouncers their entire lives. (Even if its a toddler).
Voice Recognition
“Siri, do this!” or “Alexa, do that” or “Chat-GPT, validate my life choices”, or whatever commands you give to your voiced AI is made possible thanks to supervised learning models, trained on voice samples.
Personalized Recommendations
Ever find yourself binge watching some Netflix series that showed up in your recommendations? Blame Supervised Learning, for keeping you glued to the screen.
⭐ Summary
There you have it — Supervised Learning in a nutshell: the “study with the answers” method of Machine Learning. Its how we train models to recognize patterns, make predictions, and (hopefully) not embarrass us in front of our data.
Whether it’s spam detection, price prediction, or making your computer a slightly smarter guesser, supervised learning is the trusty sidekick you’ll call on often.
But don’t worry the fun’s just getting started.
In Part 2 (you didn’t seriously think I could cover everything about this beast in one email?), we will delve deeper into the algorithms that make supervised learning tick: from straightforward Decision Trees to the more enigmatic Support Vector Machines.
Until then, I’ll catch you at our next data cycle.
