Learning boils down to looking at past data and predicting its future values in a meaningful way. When the data is continuous, like stock prices or call volumes in a call center, we call that prediction regression. If we’re predicting specific discrete classes of things like whether an image is a dog or a bird or a cat, we call that prediction classification. Classification doesn’t work only on images; you can classify all sorts of things, like text, including deciding whether the text has a positive or negative sentiment.
Sometimes, you simply want natural patterns to emerge in the data that allow you to group it, such as cluster data with related properties (such as all the coughing sounds in a big tranche of audio files) or even cell phone data that gives some hints about what types of activities its owner was doing—walking, talking, and the like.