The Machine Learning Paradigm: From Rules to Representations

In traditional software engineering, developers write explicit logic and feed input data to produce outputs:

$\text{Data} + \text{Rules} \longrightarrow \text{Answers}$

In Machine Learning, we invert this paradigm:

$\text{Data} + \text{Answers} \longrightarrow \text{Rules (Learned Model)}$

The objective is not merely to memorize training answers, but to capture the underlying probability distribution so the model can generalize to unseen samples drawn from the same domain.