Contrastive Methods
Contrastive methods learn representations by contrasting positive and negative examples. This is one of methods used in Self-supervised Learning.
The goal is to learn an encoder \(f\) such that:
\begin{equation} \text{score}(f(x), f(x^{+})) » \text{score}(f(x), f(x^{-})) \end{equation}
where \(x^{+}\) is a positive example, and \(x^{-}\) is a negative example.