CS B 553: A LGORITHMS FOR O PTIMIZATION AND L EARNING Bayesian Networks.

CS B553: ALGORITHMS FOR OPTIMIZATION AND LEARNINGBayesian Networks

AGENDA

Bayesian networks Chain rule for Bayes nets Naïve Bayes models

Independence declarations D-separation

Probabilistic inference queries

PURPOSES OF BAYESIAN NETWORKS

Efficient and intuitive modeling of complex causal interactions

Compact representation of joint distributions O(n) rather than O(2n)

Algorithms for efficient inference with given evidence (more on this next time)

INDEPENDENCE OF RANDOM VARIABLES

Two random variables a and b are independent if

P(A,B) = P(A) P(B)

hence P(A|B) = P(A) Knowing b doesn’t give you any information

about a

[This equality has to hold for all combinations of values that A and B can take on, i.e., all events A=a and B=b are independent]

SIGNIFICANCE OF INDEPENDENCE

If A and B are independent, then P(A,B) = P(A) P(B)

=> The joint distribution over A and B can be defined as a product over the distribution of A and the distribution of B

=> Store two much smaller probability tables rather than a large probability table over all combinations of A and B

CONDITIONAL INDEPENDENCE

Two random variables a and b are conditionally independent given C, if

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

hence P(A|B,C) = P(A|C) Once you know C, learning B doesn’t give

you any information about A

[again, this has to hold for all combinations of values that A,B,C can take on]

SIGNIFICANCE OF CONDITIONAL INDEPENDENCE

Consider Grade(CS101), Intelligence, and SAT Ostensibly, the grade in a course doesn’t

have a direct relationship with SAT scores but good students are more likely to get good

SAT scores, so they are not independent… It is reasonable to believe that Grade(CS101)

and SAT are conditionally independent given Intelligence

BAYESIAN NETWORK Explicitly represent independence among

propositions Notice that Intelligence is the “cause” of both Grade

and SAT, and the causality is represented explicitly

Intel.

Grade

P(I=x)

high 0.3

low 0.7

SAT

6 probabilities, instead of 11

P(I,G,S) = P(G,S|I) P(I) = P(G|I) P(S|I) P(I)

P(G=x|I) I=low I=high

‘a’ 0.2 0.74

‘b’ 0.34 0.17

‘C’ 0.46 0.09

P(S=x|I) I=low I=high

low 0.95 0.05

high 0.2 0.8

DEFINITION: BAYESIAN NETWORK

Set of random variables X={X1,…,Xn} with domains Val(X1),…,Val(Xn)

Each node has a set of parents PaX

Graph must be a DAG Each node also maintains a conditional

probability distribution (often, a table) P(X|PaX) 2k-1 entries for binary valued variables

Overall: O(n2k) storage for binary variables

Encodes the joint probability over X1,…,Xn

CALCULATION OF JOINT PROBABILITY

B E P(a|…)

TTFF

TFTF

0.950.940.290.001

Burglary Earthquake

Alarm

MaryCallsJohnCalls

P(b)

0.001

P(e)

0.002

A P(j|…)

TF

0.900.05

A P(m|…)

TF

0.700.01

P(jmabe) = ??

P(jmabe)= P(jm|a,b,e) P(abe)= P(j|a,b,e) P(m|a,b,e) P(abe)(J and M are independent given A)

P(j|a,b,e) = P(j|a)(J and B and J and E are independent given A)

P(m|a,b,e) = P(m|a) P(abe) = P(a|b,e) P(b|e) P(e)

= P(a|b,e) P(b) P(e)(B and E are independent)

P(jmabe) = P(j|a)P(m|a)P(a|b,e)P(b)P(e)

Burglary Earthquake

Alarm

MaryCallsJohnCalls

CALCULATION OF JOINT PROBABILITY

B E P(a|…)

TTFF

TFTF

0.950.940.290.001

Burglary Earthquake

alarm

MaryCallsJohnCalls

P(b)

0.001

P(e)

0.002

A P(j|…)

TF

0.900.05

A P(m|…)

TF

0.700.01

P(jmabe)= P(j|a)P(m|a)P(a|b,e)P(b)P(e)= 0.9 x 0.7 x 0.001 x 0.999 x 0.998= 0.00062

CALCULATION OF JOINT PROBABILITY

b e P(a|…)

TTFF

TFTF

0.950.940.290.001

Burglary Earthquake

alarm

maryCallsjohnCalls

P(b)

0.001

P(e)

0.002

a P(j|…)

TF

0.900.05

a P(m|…)

TF

0.700.01

P(jmabe)= P(j|a)P(m|a)P(a|b,e)P(b)P(e)= 0.9 x 0.7 x 0.001 x 0.999 x 0.998= 0.00062

P(x1x2…xn) = Pi=1,…,nP(xi|paXi)

full joint distribution

CHAIN RULE FOR BAYES NETS

Joint distribution is a product of all CPTs

P(X1,X2,…,Xn) = Pi=1,…,nP(Xi|PaXi)

EXAMPLE: NAÏVE BAYES MODELS

P(Cause,Effect1,…,Effectn)

= P(Cause) Pi P(Effecti | Cause)

Cause

Effect1 Effect2 Effectn

ADVANTAGES OF BAYES NETS (AND OTHER GRAPHICAL MODELS)

More manageable # of parameters to set and store

Incremental modeling Explicit encoding of independence

assumptions Efficient inference techniques

ARCS DO NOT NECESSARILY ENCODE CAUSALITY

A

B

C

C

B

A

2 BN’s with the same expressive power, and a 3rd with greater power (exercise)

C

B

A

READING OFF INDEPENDENCE RELATIONSHIPS

Given B, does the value of A affect the probability of C? P(C|B,A) = P(C|B)?

No! C parent’s (B) are

given, and so it is independent of its non-descendents (A)

Independence is symmetric:C A | B => A C | B

A

B

C

BASIC RULE

A node is independent of its non-descendants given its parents (and given nothing else)

WHAT DOES THE BN ENCODE?

Burglary EarthquakeJohnCalls MaryCalls | AlarmJohnCalls Burglary | AlarmJohnCalls Earthquake | AlarmMaryCalls Burglary | AlarmMaryCalls Earthquake | Alarm

Burglary Earthquake

Alarm

MaryCallsJohnCalls

A node is independent of its non-descendents, given its parents

READING OFF INDEPENDENCE RELATIONSHIPS

How about Burglary Earthquake | Alarm ? No! Why?

Burglary Earthquake

Alarm

MaryCallsJohnCalls

READING OFF INDEPENDENCE RELATIONSHIPS

How about Burglary Earthquake | Alarm ? No! Why? P(BE|A) = P(A|B,E)P(BE)/P(A) = 0.00075 P(B|A)P(E|A) = 0.086

Burglary Earthquake

Alarm

MaryCallsJohnCalls

READING OFF INDEPENDENCE RELATIONSHIPS

How about Burglary Earthquake | JohnCalls? No! Why? Knowing JohnCalls affects the probability of

Alarm, which makes Burglary and Earthquake dependent

Burglary Earthquake

Alarm

MaryCallsJohnCalls

INDEPENDENCE RELATIONSHIPS

For polytrees, there exists a unique undirected path between A and B. For each node on the path: Evidence on the directed road XEY or XEY

makes X and Y independent Evidence on an XEY makes descendants

independent Evidence on a “V” node, or below the V:

XEY, or XWY with W… Emakes the X and Y dependent (otherwise they are independent)

GENERAL CASE

Formal property in general case: D-separation : the above properties hold for all

(acyclic) paths between A and B D-separation independence

That is, we can’t read off any more independence relationships from the graph than those that are encoded in D-separation The CPTs may indeed encode additional

independences

PROBABILITY QUERIES

Given: some probabilistic model over variables X

Find: distribution over YX given evidence E=e for some subset E X / Y P(Y|E=e)

Inference problem

ANSWERING INFERENCE PROBLEMS WITH THE JOINT DISTRIBUTION Easiest case: Y=X/E

P(Y|E=e) = P(Y,e)/P(e) Denominator makes the probabilities sum to 1

Determine P(e) by marginalizing: P(e) = Sy P(Y=y,e)

Otherwise, let Z=X/(EY) P(Y|E=e) = Sz P(Y,Z=z,e) /P(e)

P(e) = Sy Sz P(Y=y,Z=z,e)

Inference with joint distribution: O(2|X/E|) for binary variables

NAÏVE BAYES CLASSIFIER

P(Class,Feature1,…,Featuren)

= P(Class) Pi P(Featurei | Class)

Class

Feature1 Feature2 Featuren

P(C|F1,….,Fn) = P(C,F1,….,Fn)/P(F1,….,Fn)

= 1/Z P(C) Pi P(Fi|C)

Given features, what class?

Spam / Not Spam

English / French / Latin

…

Word occurrences

NAÏVE BAYES CLASSIFIER

P(Class,Feature1,…,Featuren)

= P(Class) Pi P(Featurei | Class)

P(C|F1,….,Fk) = 1/Z P(C,F1,….,Fk)

= 1/Z Sfk+1…fn P(C,F1,….,Fk,fk+1,…fn)

= 1/Z P(C) Sfk+1…fn Pi=1…k P(Fi|C) Pj=k+1…n P(fj|C)

= 1/Z P(C) Pi=1…k P(Fi|C) Pj=k+1…n Sfj P(fj|C)

= 1/Z P(C) Pi=1…k P(Fi|C)

Given some features, what is the distribution over class?

FOR GENERAL QUERIES

For BNs and queries in general, it’s not that simple… more in later lectures.

Next class: skim 5.1-3, begin reading 9.1-4