Wednesday, November 28, 2012

Bayesian Networks

A Bayesian network, Bayes network, belief network, Bayes(ian) model or probabilistic directed acyclic graphical model is a probabilistic graphical model

It represents a set of random variables and their conditional dependencies via DAG
The Variables can be
1. Observable (or)
2. Latent (inferred)

Each node/variable in the network is associated with a probability function that takes as input a particular set of values for the parent variables and gives the probability of the variable represented by the node. For example, if the parents are m Boolean variables then the probability function could be represented by a table of 2^m entries, one entry for each of the 2^m possible combinations of its parents being true or false.


Parameter learning

In order to fully specify the Bayesian network and thus fully represent the joint probability distribution, it is necessary to specify for each node X the probability distribution for X conditional upon X's parents. The distribution of X conditional upon its parents may have any form. It is common to work with discrete or Gaussian distributions since that simplifies calculations.

An example Bayesian Network with parameters:

No comments:

Post a Comment