Dichotomies: Indicates that there are two categories in the classification task, for example, we want to identify whether a picture is a cat. That is to say, train a classifier, input a picture, represented by the feature vector X, the output is cat, represented by y=0 or 1. The binary classification assumes that each sample is set with only one label, 0 or 1.
Multiclass classification: Indicates that there are multiple categories in the classification task, such as sorting a bunch of fruit pictures, they may be oranges, apples, pears, etc. Multi-category classification assumes that each sample is assigned one and only one label: a fruit can be apple or pear, but not both.
Multilabel classification: Give each sample a series of target labels. Think of a data point whose attributes are not mutually exclusive (a fruit that is both apple and pear is mutually exclusive), such as a document-related topic. A text may be considered to be related to religion, politics, finance or education at the same time.