Causal and Bayesian Network(Chapter 2)

Book: Bayesian Networks and Decision Graphs

Author: Finn V. Jensen, Thomas D. Nielsen

CSE 655 Probabilistic ReasoningFaculty of Computer Science,

Institute of Business Administration

Presented byQuratulain

Quratulain 2

OutlineReasoning under uncertainty

Causal network and d-separation

Bayesian networkGraphical model

9/16/2009

Quratulain 3

Reasoning Under Uncertainty

Why Reason Probabilistically?In many problem domains it isn't possible to

create complete, consistent models of the world.

If information is given with certainty then Propositional logic (Truth table) can be used.

Want to make rational decisions even when there is not enough information to prove that an action will work.

To deal with uncertain events, we extend truth value of propositional logic to certainties which are number between 0 and 1.

9/16/2009

4

Example (Type of reasoning that human do

daily)

“In the morning, my car will not start.”Reasons:

◦I can here starter tune, so must be power in battery

◦May be fuel has been stolen overnight◦The spark plug are dirty◦May be due to the dirt in carburetor◦A loose connection in ignition system or

any thing serious

9/16/2009 Quratulain

Quratulain 5

A Causal Perspective – Car Example Construct a graph to represent

causal relationship between events which gives structure to the situation for reasoning.

9/16/2009

Variable (node)

States

Fuel {yes, no}

CleanSparkPlugs

{yes, no}

Fuel Meter {full, half , empty}

Start {yes, no}

Quratulain 6

OutlineReasoning under uncertaintyCausal network and d-separation

Bayesian networkGraphical model

9/16/2009

Quratulain 7

Causal network and d-separation

A causal network consists of a set of variables and a set of directed links between variables.

Mathematically, the structure is called a directed graph.

Causal networks can be used to follow how a change of certainty in one variable may change the certainty for other variables.

9/16/2009

Quratulain 8

3-Cases of evidence transmitionSerial Connections

Diverging Connections (common cause)

Converging Connection (common effect)

9/16/2009

P(C|A^B)=P(C|B)

P(C|A^B)=P(C|B)

Quratulain 9

Serial ConnectionsEvidence about A will influence the

certainty of B, which then influences the certainty of C.

Similarly, evidence about C will influence the certainty of A through B.

If the state of B is known, then the channel is blocked, A and C become independent.

we say that A and C are d-separated given B.

9/16/2009

Quratulain 10

Diverging Connections Influence can pass between all

the children of A if A is not known. That is, B,C, . . . , E are d-separated given A.

9/16/2009

Sex (male, female),length of hair (long, short), and stature (<168 cm, ≥168 cm)

Quratulain 11

Converging ConnectionIf nothing is known about A then

the parents are independent evidence about one of them cannot influence the certainties of the others through A.

9/16/2009

Quratulain 12

D-separationTwo distinct variables A and B in a

causal network are d-separated such that either:◦The connection is serial or diverging

and V is instantiated.◦The connection is converging, and

neither V nor any of V ’s descendants have received evidence.

9/16/2009

Quratulain 13

ExampleAre B and C independent given

A?

Are B and C independent given F

9/16/2009

Quratulain 14

Markov Blanket

The Markov blanket of a variable A is the set consisting of:◦the parents of A, ◦the children of A, and ◦the variables sharing a child with A.

The Markov blanket has the property that when instantiated, A is d-separated from the rest of the network. 9/16/2009

Quratulain 15

OutlineReasoning under uncertaintyCausal network and d-separation

Bayesian networkGraphical model

9/16/2009

16

Bayesian NetworkA Bayesian network consists of the

following◦A set of variables and a set of directed

edges between variables.◦Each variable has a finite set of

mutually exclusive states.◦The variables together with the directed

edges form an acyclic directed graph.◦To each variable A with parents B1, . . . ,

Bn, a conditional probability table P(A|B1, . . . , Bn) is attached.

9/16/2009 Quratulain

Quratulain 17

Bayesian Network The probabilities to specify are: P(A), P(B), P(C | A,B), P(E |C), P(D|C), P(F |E), and P(G| D,E,F)

It has been claimed that prior probabilities are bias to the model

Prior probabilities are necessary because prior certainty assessments are an integral part of human reasoning about certainty

The model should not include conditional independences that do not hold in the real world.

The d-separation properties check’s Conditional independences in model.

9/16/2009

Quratulain 18

Chain Rule for Bayesian NetworkLet BN be a Bayesian network

over U = {A1, . ..,An}. Then BN specifies a unique joint probability distribution P(U) given by the product of all conditional probability tables specified in BN:

9/16/2009

Quratulain 19

OutlineReasoning under uncertainty

Causal network and d-separation

Bayesian networkGraphical model

9/16/2009

Quratulain 20

Graphical ModelGraphical specification is easy for

humans to read, and helps focus attention.

The basic property of the Bayesian networks is the chain rule for compact representation of joint probability distribution.

Graphical model represents a causal relation in a knowledge domain.

9/16/2009

Quratulain 21

Questions

9/16/2009