Experts

Context

When machine learning a Bayesian network, BayesiaLab estimates probabilities using the Maximum Likelihood Estimation (MLE) method, in which the probability of each state corresponds to its observed frequency in the data set.
As such, Maximum Likelihood Estimation (MLE) is a purely Frequentist approach.
However, it is possible to introduce a Bayesian approach with this methodology by mixing the observed particles from the dataset with virtual particles defined by Dirichlet Priors.
The easiest way to define Dirichlet Priors in BayesiaLab is to specify so-called uninformative priors.
For this purpose, Uniform Prior Samples are evenly distributed across all cells of all Probability Tables and Conditional Probability Tables.
Uniform Prior Samples are a way to specify a specific kind of prior knowledge, namely the belief that all nodes are marginally independent, which would be the equivalent of a fully unconnected network.
In earlier versions of BayesiaLab, this function was known as Smoothed Probability Estimation.

To illustrate the use of Uniform Prior Samples, we use a simple network that represents the probabilistic relationship between Eye Color and Hair Color that was learned from a relatively small population of 110 subjects.
To apply Uniform Prior Samples, select Main Menu > Edit > Edit Number of Uniform Prior Samples
Then, specify the number of Prior Samples.
Setting Uniform Prior Samples to 1, means that one virtual particle (or virtual observation/occurrence) would be "spread across" all states of all nodes, assigning a fraction of this virtual particle to all cells.
For instance, in the original data, there was no observation for blond Hair Color given brown Eye Color.
Upon applying the Uniform Prior Sample of one virtual particle, the corresponding cell no longer has 0 observations.

This means that "nothing is impossible." In other words, there is no cell in any Probability Table or Conditional Probability Table in the network that contains zero observations correspond that would lead to a zero probability.
As a result, the network would no longer infer that it is impossible for a person with brown eyes to have blond hair.