Given several objects/samples x1, … , xn, the goal is to learn a hidden map h(x) that can uncover a hidden structure in the data. This hidden map can be used to ‘compress’ x (aka dimensionality reduction) or to assign to every xi a group ck (aka clustering or topic modelling). In other words, unsupervised learning does not use labelled data, and the model instead recognises patterns in the data to compute an output.
