latentspace.

standardization


Distribution standardization is the statistical process of transforming data to have a mean of $0$ and a standard deviation of $1$, often called Z-score normalization. $$ Z = \frac{X - \mu}{\sigma} $$ It converts different normal distributions into a single, standard normal distribution $Z \sim \mathcal{N}(0, 1)$ , enabling direct comparison of datasets, simplifying probability calculations, and optimizing machine learning algorithms. Based on the properties of the expected value and variance, $E(aX + b) = aE(X) + b $ and $Var(aX + b) = a^2Var(x)$, we can derive the following: $$ \begin{aligned} E(Z) &= E(\frac{1}{\sigma}X - \frac{1}{\sigma}\mu) \,\\ \,\\ &= \frac{1}{\sigma}E(X) - \frac{1}{\sigma}\mu \,\\ \,\\ &= \frac{1}{\sigma}\mu - \frac{1}{\sigma}\mu \,\\ \,\\ &= 0 \,\\ \,\\ Var(Z) &= Var(\frac{1}{\sigma}X - \frac{1}{\sigma}\mu) \,\\ \,\\ &= \frac{1}{\sigma^{\small{2}}}Var(X) \,\\ \,\\ &= \frac{1}{\sigma^{\small{2}}}\sigma^{2} \,\\ \,\\ &= 1 \end{aligned} $$