1. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 1 Jeff A. Bilmes University of Washington Department of Electrical Engineering EE512 Spring, 2006 Graphical Models Jeff A. Bilmes Lecture 1 Slides March 28 th, 2006

2. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 2 Class overview What are graphical models Semantics of Bayesian networks Outline of Today’s Lecture

3. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 3 Books and Sources for Today Jordan: Chapters 1 and 2

4. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 4 L1: Tues, 3/28: Overview, GMs, Intro BNs. L2: Thur, 3/30 L3: Tues, 4/4 L4: Thur, 4/6 L5: Tue, 4/11 L6: Thur, 4/13 L7: Tues, 4/18 L8: Thur, 4/20 L9: Tue, 4/25 L10: Thur, 4/27 L11: Tues, 5/2 L12: Thur, 5/4 L13: Tues, 5/9 L14: Thur, 5/11 L15: Tue, 5/16 L16: Thur, 5/18 L17: Tues, 5/23 L18: Thur, 5/25 L19: Tue, 5/30 L20: Thur, 6/1: final presentations Class Road Map

5. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 5 READING: Chapter 1,2 in Jordan’s book (pick up book from basement of communications copy center). List handout: name, department, and email List handout: regular makeup slot, and discussion section Syllabus Course web page: http://ssli.ee.washington.edu/ee512http://ssli.ee.washington.edu/ee512 Goal: powerpoint slides this quarter, they will be on the web page after lecture (but not before as you are getting them hot off the press). Announcements

6. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 6 A graphical model is a visual, abstract, and mathematically formal description of properties of families of probability distributions (densities, mass functions) There are many different types of Graphical model, ex: –Bayesian Networks –Markov Random Fields –Factor Graph –Chain Graph Graphical Models

7. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 7 Graphical Models Chain Graphs Causal Models DGMs UGMs Bayesian Networks MRFs Gibbs/Boltzman Distributions DBNs Mixture Models Decision Trees Simple Models PCA LDA HMM Factorial HMM/Mixed Memory Markov Models BMMs Kalman Other Semantics FST Dependency Networks Segment Models AR ZMs Factor Graphs GMs cover many well-known methods

8. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 8 Structure Algorithms Language Approximations Data-Bases Graphical Models Provide:

9. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 9 GMs give us: I.Structure: A method to explore the structure of “natural” phenomena (causal vs. correlated relations, properties of natural signals and scenes) II.Algorithms: A set of algorithms that provide “efficient” probabilistic inference and statistical decision making III.Language: A mathematically formal, abstract, visual language with which to efficiently discuss families of probabilistic models and their properties. Graphical Models Provide

10. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 10 GMs give us (cont): IV.Approximation: Methods to explore systems of approximation and their implications. E.g., what are the consequences of a (perhaps known to be) wrong assumption? V.Data-base: Provide a probabilistic “data-base” and corresponding “search algorithms” for making queries about properties in such model families. GMs Provide

11. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 11 GMs There are many different types of GM. Each GM has its semantics A GM (under the current semantics) is really a set of constraints. The GM represents all probability distributions that obey these constraints, including those that obey additional constraints (but not including those that obey fewer constraints). Most often, the constraints are some form of factorization property, e.g., f() factorizes (is composed of a product of factors of subsets of arguments).

12. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 12 Types of Queries Several types of queries we may be interested in: –Compute: p(one subset of vars) –Compute: p(one subset of vars| another subset of vars) –Find the N most probable configurations of one subset of variables given assignments of values to some other sets –Q: Is one subset independent of another subset? –Q: Is one subset independent of another given a third? How efficiently can we do this? Can this question be answered? What if it is too costly, can we approximate, and if so, how well? These are questions we will answer this term. GMs are like a probabilistic data-base (or data structure), a system that can be queried to provide answers to these sorts of questions.

13. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 13 Example Typical goal of pattern recognition: –training (say, EM or gradient descent), need query of form: In this form, we need to compute p(o,h) efficiently. –Bayes decision rule, need to find best class for a given unknown pattern: –but this is yet another query on a probability distribution. –We can train, and perform Bayes decision theory quickly if we can compute with probabilities quickly. Graphical models provide a way to reason about, and understand when this is possible, and if not, how to reasonably approximate.

14. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 14 Some Notation Random variables X i, Y i, X, Y (scalar or vector) Distributions: Subsets:

15. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 15 Main types of Graphical Models Markov Random Fields –a form of undirected graphical model –relatively simple to understand their semantics –also, log-linear models, Gibbs distributions, Boltzman distributions, many “exponential models”, conditional random fields (CRFs), etc. Bayesian networks –a form of directed graphical model –originally developed to represent a form of causality, but not ideal for that (they still represent factorization) –Semantics more interesting (but trickier) than MRFs Factor Graphs –pure, the assembly language models for factorization properties –came out of coding theory community (LDPC, Turbo codes)

16. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 16 Main types of Graphical Models Chain graphs: –Hybrid between Bayesian networks and MRFs –A set of clusters of undirected nodes connected as directed links –Not as widely used, but very powerful. Ancestral graphs –we probably won’t cover these.

17. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 17 C X Mixture models Q1Q1 Q2Q2 Q3Q3 Q4Q4 Markov Chains Q1Q1 Q2Q2 Q3Q3 Q4Q4 Bayesian Network Examples

18. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 18 X Y Other generalizations possible E.g., Q = gen. diagonal, or capture using general A since  Q R PCA: Q = , R  0, A = ortho FA: Q = I, R = diagonal GMs: PCA and Factor Analysis

19. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 19 X1X1 X2X2 X3X3 X4X4 I1I1 I2I2 The data X 1:4 is explained by the two (marginally) independent causes. Independent Component Analysis

20. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 20 C X   Class conditional data has diff. mean but common covariance matrix. Fisher’s formulation: project onto space spanned by the means. Linear Discriminant Analysis

21. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 21 C X   C X   M C X   M HDA/QDA MDA HMDA Extensions to LDA Heteroschedastic Discriminant Analysis/Quadratic Discriminant Analysis Mixture Discriminant Analsysis Both

22. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 22 Generalized Probabilistic Decision Trees Hierarchical Mixtures of Experts (Jordan) O I D1D1 D2D2 D3D3 Generalized Decision Trees

23. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 23 Example: Printer Troubleshooting Dechter

24. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 24 Classifier Combination X S C Mixture of Experts (Sum rule) X1X1 C Naive Bayes (prod. rule) X2X2 XNXN... Other Combination Schemes C C1C1 C2C2 X1X1 X2X2

25. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 25 C X C X Generative Model Discriminative Model Discriminative and Generative Models

26. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 26 HMMs/Kalman Filter Q1Q1 Q2Q2 Q3Q3 Q4Q4 X1X1 X2X2 X3X3 X4X4 Q1Q1 Q2Q2 Q3Q3 Q4Q4 X1X1 X2X2 X3X3 X4X4 HMM Autoregressive HMM

27. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 27 Switching Kalman Filter S1S1 S2S2 S3S3 S4S4 X1X1 X2X2 X3X3 X4X4 Q1Q1 Q2Q2 Q3Q3 Q4Q4

28. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 28 Factorial HMM Q1Q1 Q2Q2 Q3Q3 Q4Q4 X1X1 X2X2 X3X3 X4X4 Q’ 1 Q’ 2 Q’ 3 Q’ 4

29. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 29 Example: standard 4-gram Standard Language Modeling

30. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 30 Nothing gets zero probability Interpolated Uni-,Bi-,Tri-Grams

31. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 31 Conditional mixture tri-grams

32. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 32 Bayesian Networks … and so on We need to be more formal about what BNs mean. In the rest of this, and in the next, lecture, we start with basic semantics of BNs, move on to undirected models, and then come back to BNs again to clean up …

33. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 33 Bayesian Networks Has nothing to do with “Bayesian statistical models” (there are Bayesian and non-Bayesian Bayesian networks). valid:invalid:

34. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 34 Sub-family specification: Directed acyclic graph (DAG) Nodes - random variables Edges - direct “influence” Instantiation: Set of conditional probability distributions 0.90.1 e b e 0.20.8 0.01 0.99 0.90.1 be b b e BE P(A | E,B) Earthquake Radio Burglary Alarm Call Compact representation: factors of probabilities of children given parents (J. Pearl, 1988) Together (graph and inst.): Defines a unique distribution in a factored form BN: Alarm Network Example

35. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 35 Bayesian Networks Reminder: chain rule of probability. For any order 

36. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 36 Bayesian Networks

37. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 37 Conditional Independence

38. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 38 Conditional Independence & BNs GMs can help answer CI queries: Y D CE BZX A No!!

39. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 39 Bayesian Networks & CI

40. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 40 Bayesian Networks & CI

41. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 41 Bayesian Networks & CI

42. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 42 Bayesian Networks & CI

43. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 43 Pictorial d-separation: blocked/unblocked paths Blocked Paths Unblocked Paths

44. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 44 Three 3-node examples of BNs and their conditional independence statements. V1V1 V2V2 V3V3 V1V1 V2V2 V3V3 V1V1 V2V2 V3V3 Three Canonical Cases

45. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 45 Case 1 Markov Chain

46. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 46 Case 2 Still a Markov Chain

47. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 47 Case 3 NOT a Markov Chain

48. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 48 SUVs Greenhouse Gasses Global Warming Lung Cancer Smoking Bad Breath GeneticsCancerSmoking Examples of the three cases

49. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 49 Bayes Ball simple algorithm, ball bouncing along path in graph, to tell if path is blocked or not (more details in text).

50. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 50 What are implied conditional independences?

51. Lec 1: March 28th, 2006EE512 - Graphical Models - J. BilmesPage 51 Two Views of a Family