Introduction to Bayesian Networks

The approach

Another approach for creating structured probabilistic models for plan recognition employs BNs. BNs are a type of probabilistic graphical model that represent the probabilistic relationships between variables as a directed graph. The nodes within the graph contain conditional probability tables for a variable based on the information flow into that node from other nodes in the network (the parent variables). BNs compactly and efficiently represent a probabilistic model because they use Bayes’ rule and conditional probability to avoid storing the entire joint probability distribution for all the variables.

What are Bayesian networks (BNs)?

Bayesian networks (BNs) (Pearl, 1988) are a probabilistic graphical representation of variables and their dependencies. Imagine wanting to deduce if a player is highly ranked or not. To deduce this, four variables come into play, as shown in the figure. High rank depends on the variable score, which in turn depends on both the champion and the role. The probability of a high rank depends on the score.