Real-life example.

Hi Louisa!   1) Supervised learning would be any linear regression, logistic regression, neural network, etc. where you have past data about the inputs and the outcomes. So, you know if someone bought your product, if someone subscribed, if someone clicked a button, liked a post, moreover how much was spent, how many miles were ran, etc. Each of those represent an outcome which you know happened.   This is what supervised learning is about - you know the factors, you know the result. Now, based on this, you are trying to find out how these factors, translate into these results.   Hence supervised -> we are supervising by giving the correct output.   2) Unsupervised learning is whenever you don't have an outcome from past data. Examples are cluster analysis, factor analysis, PCA, etc. For instance, you know a bunch of people came to your website, and are trying to find out how are they similar. You don't have an answer for that. You know you can group them in some way, but there is no 1 way to do that. With unsupervised learning you are saying: these people, images, observations can be grouped in some way. I don't know what way, but you, algorithm, should find a way to do that.   For images that may be: dogs have a different kind of fur than cats, their eyes are different, their noses, paws, etc. are different. So the algorithm will probably group them in this way.   Or people can be coming from USA, India, UK, Australia, China, etc. So the algorithm will group them based on their country of origin, language, etc.   You don't know what's the best way prior to clustering, so it is unsupervised.   Another example is grouping features (not observations). So, if you have 1000 input features, you can group them in 5 big groups for instance and run an algorithm with 5 input features. The process of creating these groups would be a prime example of unsupervised learning.   3) Reinforcement learning is very similar to supervised learning, but is not the same. Your 'target' is to get constantly better at what you are doing (playing chess, Go, driving a car, walking, etc.). So you don't have a target: 0 or 1, because driving a car is not 1 single outcome. It is a continuous collection of many, many correct decisions one after the other with no option to do it wrong. So it is much harder than supervised learning, where 1 mistake doesn't mean much if you got 10,000,000 more correctly.   Hope this helps! Best,
The 365 Team