Probabilistic Graphical Models
Cmput 651; Sept-Dec 2008
Computing Science, University of Alberta

WARNING: This material is still being tweaked!
The symbol means the material is basically ready to print.
See CG schedule for CG's lecture notes.

 

Material covered

Slides

Readings / References
Module 0: Introduction
  L#1-2  

KF: 1, 2.1, 2.3?  (J: 3.1)
Graphical models. M. Jordan. Statistical Science, 2004
Cribsheet (I Murray, 2003)

Module 1: Directed Graphical Models
 Bayesian networks: representation/semantics, learning, inference
  L#3 - 5
  • Belief Net Representation / Semantics
    		•  KF: 3
    Intro Bayesian Net
    Local Belief Net
      L#6 - 8
  • Learning Belief Net Parameters: Complete Data
       Framework: Frequentist, Bayesian
    		• KF: 15, 16
      L#9
  • Learning Belief Net Parameters: Partial Data
       Framework: Frequentist
    		• KF: 18--18.2
      L#10
  • Learning Belief Net Structures - complete data
       Constraint-based; Chow-Liu; Fixed-order; Structure search
    		• KF: 17
      L# 11
  • Exact inference:
       Variable elimination
    		• KF: 8.1-8.3; 8.7.1
    Module 2: Undirected Graphical Models
     Markov random fields, Factor graphs
      L#12-14
  • Semantics of undirected models
  • Directed vs. undirected models
  • Clique trees
    		•  KF: 4 - 4.4, 4.6
    KF: 4.5
    KF: 9, Wikipedia: Junction Tree Alg
    Module 3: Approximate Inference
      L#15
    Global Approximate Inference: Inference as Optimization
    • Propagation based approximation
    • Loopy belief propagation
    • Structured variational approximations
    		KF: 10.1-3, 10.5
      L#16-17
    (31 Oct, 3 Nov)
    Lay of the Land Presentations

      L#18
    Particle-Based Approximate Inference
    • Complete Particles ... Importance Sampling
    • Markov Chains; Monte Carlo; Gibbs Sampling (balance eq), Metropolis-Hastings
    		 KF: 11.1-4
    Module 4: Learning Revisited
      L#19
    • Learning Undirected Models
    		KF: 19,   (J: 9, 20)
    Jordan et al., "Thin Junction Trees"
    S. Della-Pietra, V. Della-Pietra and J. Lafferty, "Inducing Features of Random Fields"
      L#20

    • Partially Observed Data
    		 KF: 18,   KF: 19.3.3
    Other slides
    Module 5: Gaussian and Hybrid Models
      L#21
      L#22

    • Gaussian (continuous) Models
    • Inference in Hybrid Networks
    		KF: 5.5, KF: 6
    KF: 5.5.1, KF Hybrid Network Chapter handout (get at Main CS Office, 2nd floor Ath Hall)
      L#23-24
    (28 Nov, 1 Dec)    		 Final Presentations

    Some material was borrowed (ok, "stolen", but with permission) from C Guestrin, CMU --- 10708-F06