Section 4 : Dynamic Bayesian networks

Commentary

Section Goals

• To discuss dynamic Bayesian networks (DBNs), which include Hidden Markov Models and Kalman filters as special cases.

Learning Objectives

Learning Objective 1

• Describe the structure of dynamic Bayesian networks.
• Explain the differences between DBN, HMM, and Kalman filter.
• Exemplify a DBN application, and draw its structure.
• Explain the ideas behind the construction of DBNs.
• Describe both exact and approximate inference in DBNs.

Objective Readings

Required readings:

Reading topics:

Dynamic Bayesian Networks (see Section 15.5 of AIMA3ed)

Smyth, P., Heckerman, D., and Jordan, M. I. (1997). Probabilistic independence networks for hidden Markov probability models. Neural Computation, 9(2), 227-269.

Supplemental Readings

Huang, C.-L., Shih, H.-C., and Chao, C.-Y. (2006). Semantic analysis of soccer video using dynamic Bayesian network. IEEE Transactions on Multimedia, 8(4), 749 - 760. Digital Object Identifier 10.1109/TMM.2006.876289

Objective Questions

• Why and how can exact inference in DBNs be performed using the algorithms for inference in Bayesian networks?
• What is a better solution for approximate inference in DBNs?

Objective Activities

• Explore the following source code for the particle filtering algorithm, which is downloadable from
• the textbook's website.
  • Particle-Filtering
• Complete Exercise 15.17 of AIMA3ed.