Meeting 19: Bayesian Networks

Reading: AIAMA 14.1-14.4.1

Probabalistic agent models and algorithms for doing inference on them can get quite complex. A useful way to develop, visualize, describe, and represent such models is through a graphical interpretation called Bayesian Networks (BN) or Belief Networks.

Questions you should be able to answer after reading are:

  1. How does one model an uncertain environment using a BN?
  2. What do the nodes of a BN represent?
  3. What do the edges/connections in a BN represent?
  4. How do you go back and forth between a BN and the model joint distribution?
  5. Why are models resulting in sparse graphs desirable?
  6. What does conditional independence in a BN mean?
  7. What is the Markov blanket?
  8. What information and data structures are needed to represent a BN?
  9. How does one do inference using enumeration?