Unit 7: Temporal Probabilistic Reasoning and Dynamic Bayesian Networks

Commentary

In the last Unit, we discussed exact and approximate probabilistic reasoning, during which no state or environment changes are taken into account. However, agents in uncertain environments should be able to keep track of the current state of the environment, and the uncertainty of how the environment changes over time. This unit will discuss these issues by defining several basic inference tasks, by describing the general structure of inference algorithms for temporal models,and by describing three specific kinds of models: hidden Markov models, Kalman filters, and dynamic Bayesian networks. Considering that real-world application always involves temporal uncertainty in reasoning and decision-making, it is not surprising that temporal probabilistic reasoning approaches are useful in many application problems, including speech recognition, which is also introduced in this unit. Machine learning (See Units 9 - 11) plays a central role in the construction of these models.

Unit Purpose

When you complete this unit, you will be able to

• Define concepts about representing time, uncertainty, and the basic inference tasks; i.e., filtering, prediction, smoothing, and most likely explanation.
• Analyse the mechanisms and algorithms of hidden Markov models and Kalman filters.
• Explain the unified description of temporal probabilistic reasoning using the notation from dynamic Bayesian networks.
• Discuss how temporal probabilistic reasoning helps to solve speech recognition problems.

Section 1: Representation and inference in temporal models
Section 2: Hidden Markov models (HMMs)
Section 3: Kalman filters
Section 4: Dynamic Bayesian networks
Section 5: Speech Recognition as an application of HMM

Readings

Supplemental Unit Readings

Books:

Bishop, C.M. (2006). Pattern Recognition and Machine Learning. Springer. ISBN 0-387-31073-8. (Refer primarily to the sections covering Graphical Models)

Unit 7: Temporal Probabilistic Reasoning and Dynamic Bayesian Networks

Pearl, J. (1988). Probabilistic reasoning in intelligent systems: Networks of plausible inference. San Mateo, CA: Morgan Kaufmann.

Activities

- Explore other applications of HMMs, Kalman filters, and dynamic Bayesian networks in fields that are familiar to you, and report your discoveries and comments in the course conference.