# Quantum Bayesian Networks

## April 30, 2017

### I don’t use plates to eat Bayesian networks

Filed under: Uncategorized — rrtucci @ 2:02 am

I am referring here to the “plates” used by some to draw Bayesian networks. Here is why I don’t like them. Consider this diagram given as an example by the Wikipedia article on plate notation. In this diagram, I would replace
$\theta$ by $\theta^{M}$, $z$ by $z^{NM}$, and $w$ by $w^{NM}$. F*** (forget) the plates.

This aversion of mine to plates is related to the following idiosyncratic notation of mine.

In computer code, I like to use $x\_$ for a random variable x and $vx$ for a vector of $N$ observations x. (In a latex document I might use $\underline{x}$ for a random variable and $\underline{x}^N$ for the $N$ observations.)

A vector of x measurements $vx$ is like a primitive version of the random variable $x\_$. In fact, from $vx$ one can get an empirical distribution $P_{emp}(x)$ which approximates the true distribution $P(x)$ which defines the random variable $x\_$. That’s why when I see a statement like $x\_ \sim P(x)$, I think of this as an ordinary equivalence relation. In fact,

$x\_ \sim vx \sim P_{emp}(x) \sim P(x)$

are all equivalent in the limit of a large number of observations.