Supervised, Unsupervised, Semi-supervised, Self-supervised, and Reinforcement Learning are all different ways of learning from data to tackle problems.

Supervised learning is when you try to learn a model from data that is labeled, that is you know what the algorithm is supposed to output for some given inputs, and you try to learn to predict those outputs. For example, given that you have a bunch of pictures of handwritten digits, and what those digits are supposed to be, you would use supervised learning to try to take in an image and make a prediction about what the digit is in that picture. Unsupervised learning is when you are trying to learn without labels. The canonical example is clustering, where given a bunch of data, you try to group together the data based on position.

Semi-supervised learning is when you combine both supervised and unsupervised learning in order to learn as much as possible. For example, you might cluster the data together, and then use supervised learning in order to fix those clusters and to label each cluster automatically.

Self-supervised learning is when you reformulate an unsupervised learning problem as a supervised one. For example, given a sentence, you might remove a word and then try to make your model learn to predict the missing word.

Finally, reinforcement learning is where you try to learn from the environment and the feedback from it. An easy example here is video games, where going in, the model has no idea how to play, but from every movement in the game, they might get points (or die) and learn from that feedback while they play.