Machine learning comes in three primary flavours: supervised learning, unsupervised learning, and reinforcement learning. Each type has its own purpose, approach, and applications, and in this blog post I hope I can help you break them down to understand each type.

Supervised Learning

In supervised learning, our goal is straightforward. We’re looking to predict a target variable based on past observations. This approach typically covers two main tasks:

• Classification – Here, we’re assigning categories. Think of it like teaching a model to recognize certain types of objects, like distinguishing between emails that are spam or not spam.

• Regression – Instead of categories, regression focuses on predicting a continuous number. For example, it could estimate the selling price of a house based on how many bedrooms it has.

Supervised learning requires a robust dataset with both inputs and their corresponding target values. In other words, if we want our model to recognize cats, we need a labeled dataset of images—each picture tagged with “cat”. Once trained, the model should then recognize a cat, even if it’s never seen that specific cat before.

Unsupervised Learning

Unsupervised learning takes a different path. Instead of predicting a target variable, it looks for patterns or groupings within the data, BUT without the need for labeled outputs. This is where techniques like clustering and anomaly detection come into play.

A common application is market segmentation: grouping customers based on their behaviours or preferences without having predefined categories. It’s like giving our model a set of animal images without telling it what each animal is. The model might group them based on colour or shape, perhaps putting cats together with dogs, but it won’t actually “know” what an cat or a dog is.

Reinforcement Learning

Finally, reinforcement learning is all about strategy and goal oriented learning through interaction. In this approach, a model learns by making decisions and receiving feedback on its actions. Theres a youtube video of this approach being used to play Pokemon, so lets think of it like that, it’s like a player learning to improve in a game, learning from mistakes. It’s the technique often used in teaching computers to play games like chess, where each move can bring a reward (winning the game) or a penalty (losing).

Comparison

Summary

Machine Learning has a role to play in nearly every industry today, from predicting product demand to recognizing medical conditions. These foundational methods; supervised, unsupervised, and reinforcement learning are the building blocks that make it all possible. Want to read more about where ML fits within AI? Check out this other blog post.