Machine Learning Algorithms

Learnability

A parametric model can be split into two parts: a static structure and a dynamic set of parameters. The former is determined by choice of a specific algorithm and is normally immutable (except in the cases when the model provides some re-modeling functionalities), while the latter is the objective of our optimization. Considering n unbounded parameters, they generate an n-dimensional space (imposing bounds results in a sub-space without relevant changes in our discussion) where each point, together with the immutable part of the estimator function, represents a learning hypothesis H (associated with a specific set of parameters):

The goal of a parametric learning process is to find the best hypothesis whose corresponding prediction ...