Characterizing the invariances of learning algorithms using category theory

Kenneth D. Harris

Many learning algorithms have invariances: when their training data is transformed in certain ways, the function they learn transforms in a predictable manner. Here we formalize th…