Suppose you have data as follows. | | Age | Points | | ----- | --- | ------ | | Alice | $1$ | $2$ | | Bob | $3$ | $4$ | BatchNorm (resp. LayerNorm) maps each $X\in\{\{1,3\},\{2,4\}\}$ (resp. $X\in\{\{1,2\},\{3,4\}\}$) to $$ \gamma \left( \frac{x-\mu}{\sigma} \;\middle|\; x\in X \right) + \beta $$ where $\gamma\in\mathbb{R}$ is a learned scale parameter of $X$, and $\beta\in\mathbb{R}$ is a learned shift parameter of $X$, and $\mu\in\mathbb{R}$ is the mean of $X$, and $\sigma\in\mathbb{R}$ is the standard deviation of $X$.