All the lecture notes are linked to this web page, in PDF format, with 2 slides per page. The reader for PDF files is available free from Adobe for UNIX, Apple Macintosh, and Windows.
I am happy to provde source files for all these slides upon request for teaching purposes.
Introduction |
RN Chapter 13 Slides: PDF |
|||
Conditional independence. Introduction to belief networks |
RN, Sec. 14.1, 14.2; MJ, Sec. 2.1 Slides: PDF |
|||
Independence maps. Bayes ball algorithm |
Slides: PDF |
|
||
More on Bayes nets: d-separation, moral graph, practical considerations |
Slides: PDF |
|||
Undirected graphical models |
Slides: PDF |
|||
Exact inference: Variable elimination |
Slides: PDF |
|||
Exact inference: Message passing in trees. Clique trees |
Slides: PDF |
|||
Message passing in clique trees |
Slides: |
|||
Junction trees |
Slides: |
|||
Belief propagation in polytrees |
Slides: |
|||
No class |
Slides: |
|||
Loopy belief propagation |
Slides: |
|||
Approximate inference: Likelihood weighting |
Slides: |
|
||
Importance sampling and particle filters |
Slides: |
|||
No class |
Slides: |
|||
No class |
Slides: |
|||
Approximate inference: Gibbs sampling. MCMC methods |
Slides: |
|
||
Introduction to learning. Maximum likelihood |
Slides: |
|||
Bayesian learning |
Slides: |
|
||
Structure learning in Bayes nets with complete data |
Slides: |
|
||
Expectation maximization |
Slides: |
|||
More on expectation maximization. Applications to clustering |
Slides: |
|||
No class |
Slides: |
|||
Exponential family distributions |
Slides: |
|||
Learning undirected graphical models |
Slides: |
|||
Evaluation and comparison of different algorithms |
Slides: |
|||
Hidden Markov models. Forward-backward inference algorithm |
Slides: |
|||
Expectation maximization for learning HMMs. Dynamic Bayesian networks (DBN) |
Slides: |
|||
Kalman filter and related models |
Slides: |
|||
Introduction to decision making. Utility theory |
Slides: |
|
||
Markov Decision Processes. Policies and value functions. Policy evaluation |
Slides: |
|||
Optimal policies and value functions. Policy iteration and value iteration algorithms |
Slides: |
|||
Monte Carlo and temporal difference learning for policy evaluation |
Slides: |
|||
On-policy control learning: Sarsa. Exploration-exploitation trade-off |
Slides: |
|||
Off-policy learning: Policy evaluation, Q-learning |
Slides: |
|||
Actor-critic algorithms |
Slides: |
|||
Partially Observable Markov Decision Processes (POMDPs). Planning methods for POMDPs |
Slides: |
|||
More on planning for POMDPs. Learning methods for POMDPs |
Slides: |
|||
Other methods for decision making under uncertainty |
Slides: |
|