Graphical Models 4dummies

36
Graphical Models for dummies Max Khesin, Data Strategist, Liquidnet Inc.

description

 

Transcript of Graphical Models 4dummies

Page 1: Graphical Models 4dummies

Graphical Models for dummies

Max Khesin, Data Strategist, Liquidnet Inc.

Page 2: Graphical Models 4dummies

Graphical Models ForBy Dummies

Page 3: Graphical Models 4dummies

Grand Theme

• “Probabilistic graphical models are an elegant framework which combines uncertainty (probabilities) and logical structure (independence constraints) to compactlyrepresent complex, real-world phenomena”. (Koller 2007)

Page 4: Graphical Models 4dummies

Trying to guess if family is home.

• When wife leaves the house she leaves the outdoor light on (but sometimes leaves it on for a guest)

• When wife leaves the house, she usually puts the dog out

• When dog has a bowel problem, she goes to the backyard

• If the dog is in the backyard, I will probably hear it (but it might be the neighbor's dog)

Page 5: Graphical Models 4dummies
Page 6: Graphical Models 4dummies

These causal connections are not absolute

Three causes of uncertainty (Norvig, Russell 2009):

- Laziness

- Theoretical Ignorance

- Practical Ignorance

Page 7: Graphical Models 4dummies

Problem with probability

• Too many parameters

• For binary random variables, 2^n-1

Page 8: Graphical Models 4dummies

Bayesean Network - definition

• A Bayesian network is a directed graph in which each node is annotated with quantitative probability information. The full specification is as follows:

1. Each node corresponds to a random variable, which may be discrete or continuous.

2. A set of directed links or arrows connects pairs of nodes. If there is an arrow from node X to node Y , X is said to be a parent of Y. The graph has no directed cycles (and hence is a directed acyclic graph, or DAG.

3. Each node Xi has a conditional probability distribution P(Xi | Parents(Xi)) that quantifies the effect of the parents on the node. (Russell, Norvig 2009)

Page 9: Graphical Models 4dummies
Page 10: Graphical Models 4dummies

Compressed distribution (factorization)

• In our Bayesian system, we only have 10 parameters

• The compression is due to independence

• Independence is how causality manifests itself in the distribution

Page 11: Graphical Models 4dummies

D-connection

Page 12: Graphical Models 4dummies

Definition: conditional probability

Page 13: Graphical Models 4dummies

Decomposing a joint distribution

Page 14: Graphical Models 4dummies

Topologic sort:

• “topological ordering of a directed acyclic graph (DAG) is a linear ordering of its nodes in which each node comes before all nodes to which it has outbound edges” - Wikipedia.

Page 15: Graphical Models 4dummies

Sorting the family-home example

• Family-out(1), Bowel-problem(2), Ligths-On(3), Dog-out(4), Hear-bark(4)

• Bowel-problem(1), Family-out(2), Ligths-On(3), Dog-out(4), Hear-bark(5)

• Right away, we got all the non-descendants to the left of the variable

Page 16: Graphical Models 4dummies

Parents only matter

Page 17: Graphical Models 4dummies

Chain rule for Bayesian Networks

Page 18: Graphical Models 4dummies
Page 19: Graphical Models 4dummies
Page 20: Graphical Models 4dummies

Markov Networks

• These models are useful in modeling a variety of phenomena where one cannot naturally ascribe a directionality to the interaction between variables

Page 21: Graphical Models 4dummies

Markov Networks

Page 22: Graphical Models 4dummies

Markov Networks

• D is a set of random variables

• Factor to be a function from Val(D) to IR+.

Page 23: Graphical Models 4dummies

Markov Networks

• H is a Markov network structure

• set of subsets D1, . . . ,Dm, where each Di is a complete subgraph of H

• factors

Page 24: Graphical Models 4dummies

Markov Network - factorization

Where the unnormalized measure is

And normalization factor is

Page 25: Graphical Models 4dummies

Factor Product (pointwise multiplication)

Page 26: Graphical Models 4dummies
Page 27: Graphical Models 4dummies

Decision Networks

- Combine Bayesian Networks with Utility Theory

Page 28: Graphical Models 4dummies

Utility-based agent

Page 29: Graphical Models 4dummies

Decision network

Page 30: Graphical Models 4dummies

Evaluating decision network

Page 31: Graphical Models 4dummies

Applications – Bayes Nets

• Expert systems. • “…A later evaluation showed that the diagnostic accuracy of

Pathfinder IV was at least as good as that of the expert used to design the system. When used with less expert pathologists, the system significantly improved the diagnostic accuracy of the physicians alone. Moreover, the system showed greater ability to identify important findings and to integrate these findings into a correct diagnosis. Unfortunately, multiple reasons prevent the widespread adoption of Bayesian networks as anaid for medical diagnosis, including legal liability issues for misdiagnoses and incompatibility with the physicians' workflow” (Koller 2009)

Page 32: Graphical Models 4dummies

Applications – Markov Networks

• Computer vision – segmentation

• Regions are contiguous. Glove is next to the arm

“superpixels”

Page 33: Graphical Models 4dummies

Application – Markov Nets (combining logic and probability)

)()(),(,

)()(

ySmokesxSmokesyxFriendsyx

xCancerxSmokesx

1.1

5.1

Cancer(A)

Smokes(A)Friends(A,A)

Friends(B,A)

Smokes(B)

Friends(A,B)

Cancer(B)

Friends(B,B)

Two constants: Anna (A) and Bob (B)

Page 34: Graphical Models 4dummies

Road to AGI…

Page 36: Graphical Models 4dummies

References

• Russell, Norvig 2009: Artificial Intelligence: A Modern Approach (AIMA) http://aima.cs.berkeley.edu/ (amazon)

• Getoor, Taskar 2007: Introduction to Statistical Relational Learning http://www.cs.umd.edu/srl-book/ (amazon)

• Koller, Friedman 2009: Probabilistic Graphical Models: Principles and Techniques http://pgm.stanford.edu/ (amazon)

• Charniak 1991: Bayesian Networks without Tears. www.cs.ubc.ca/~murphyk/Bayes/Charniak_91.pdf

• CS228: http://www.stanford.edu/class/cs228/ (course available via SCPD)

• Domingos, practical statistical learning in AI http://www.cs.cmu.edu/~tom/10601_sp08/slides/mlns-april-28.ppt, see also http://www.youtube.com/watch?v=bW5DzNZgGxY

• Koller 2007: “Graphical Models in a Nutshell”, a chapter of Getoor, Taskar 2007, availavle online http://www.seas.upenn.edu/~taskar/pubs/gms-srl07.pdf