siamese neural networks

its a method to learn models based on two or more datapoints where one learns an embedding function \(G(\vect{y}) = \hat{\vect{y}}\) so that

\begin{equation} \label{org9207932} \mathbb{D}(\vect{x},\vect{y}) = \norm{G(\vect{x}) - G(\vect{y}))} \end{equation}

or more generally

\begin{equation} \label{org6e430a2} \mathbb{S}(\vect{x},\vect{y}) = C(G(\vect{x}),G(\vect{y})) , \end{equation}

where \(C\) is any static binary metric function.

See Mathisen2019