The medoid function is defined in this graph neural network paper as: $$ t := \arg\min_{y\in \mathcal{X}}\sum_{j=1}^N||x_j-y||$$ which is a "multivariate generalization of the Median" and $\mathcal{X}=\{x_1,x_2,...,x_n\}$ is a collection of points in $\mathbb{R}^d$.
How does this relate to the median in higher dimensions? In $\mathbb{R}^1$, the median uses an ordered list of data-points and finds the central point, what does it mean to find the median in $\mathbb{R}^2$ and beyond, and how does this function do it?
I've never seen this function before, so i wonder if it has it's drawbacks and if there are better formulas to use? Furthermore, Wikipedia states the medoid "is not equivalent to a median" so is this paper wrong?