Maximum likelihood — make the data probable
Choosing a model's parameters to make the observed data as probable as possible, in practice by minimizing the negative log-likelihood.
For a coin, the maximum-likelihood estimate is the observed frequency of heads (section 3.4).