Aiml Unit 2

CS3491– ARTIFICIALINTELLIGENCEANDMACHINELEARNING UNIT2
DEPARTMENTOFCOMPUTERSCIENCEANDENGINEERINGII
YEAR/IVSEM
CS3491–ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING
UNITII PROBABILISTICREASONING
SYLLABUS:
Acting under uncertainty – Bayesian inference – Naïve Bayes models.
Probabilistic reasoning – Bayesian networks – exact inference in BN–
approximate inference in BN–causal networks.
PARTA
1. Define uncertainty and list the causes of uncertainty.
Uncertainty:
 The knowledge representation, A→B, means if A is true then B is true, but
a situation where not sure about whether A is true or not then cannot
express this statement, this situation is called uncertainty.
 So to represent uncertain knowledge, uncertain reasoning or probabilistic
reasoning is used.
Causes of uncertainty:
1. Causes of uncertainty in the real world
2. Information occurred from unreliable sources.
3. Experimental Errors
4. Equipment fault
5. Temperature variation
6. Climate change.
2. Define Probabilistic reasoning. Mention the need of probabilistic

reasoning in AI
Probabilistic reasoning:
 Probabilistic reasoning is a way of knowledge representation, the
concept of probability is applied to indicate the uncertainty in
knowledge.
Need of probabilistic reasoning in AI:
 When there are unpredictable outcomes.
 When specifications or possibilities of predicates becomes too large to
handle.
 When an unknown error occurs during an experiment.
by Ramya p
3. List the Ways to solve problems with uncertain knowledge.

 Bayes' rule
 Bayesian Statistics
4. Define Probability and the probability of occurrence.

 Probability can be defined as a chance that an uncertain event will
occur.
 Thevalueofprobabilityalwaysremainsbetween0and1thatrepresent ideal
uncertainties.
o 0≤P(A)≤1, where P(A)is the probability of an event A.
o P(A)=0,indicates total uncertainty in an event A.
o P(A)=1,indicates total certainty in an event A.
 Formula to find the probability of an uncertain event
P(¬A) = probability of a not happening event. (¬A) + P(A) = 1.
5. Define the terms event, sample space, random variables, prior

probability and posterior probability.
 Event: Each possible outcome of a variable is called an event.
 Sample space: The collection of all possible events is called sample
space.
 Random variables: Random variables are used to represent the events
and objects in the real world.
 Prior probability: The prior probability of an event is probability
computed before observing new information.
 Posterior Probability: The probability that is calculated after all
evidence or information has taken into account. It is a combination of prior
probability and new information.
6. Define Conditional probability.

 Conditional probability is a probability of occurring an event when
another event has already happened.
 Let's suppose, to calculate the event A when event B has already
occurred, "the probability of A under the conditions of B", it is:
by Ramya p
Where P(A⋀B)= Joint probability of a and B

P(B)= Marginal probability of B.
7. In a class, there are 70% of the students who like English and 40% of
the students who likes English and mathematics, and then what
Is the percent of students
Those who like English also like mathematics?
Solution:
Let, A is an event that a student likes
Mathematics B is an event that a student
likes English.
Hence, 57%arethestudentswholikeEnglishalsolikeMathematics
8. Define Bayesian Inference.

 Bayesian inference is a probabilistic approach to machine learning that
provides estimates of the probability of specific events.
 Bayesian inference is a statistical method for understanding the
uncertainty inherent in prediction problems.
 Bayesian inference algorithm can be viewed as a Markov Chain
Monte Carlo algorithm that uses prior probability distributions to
optimize the likelihood function.
9. List Bayes Theorem or Bayes Rule

 Bayes' theorem can be derived using product rule and conditional
probability of event A with known event B:
 Product Rule:
1. P(A⋀B)=P(A|B)P(B)or
2. P(A⋀B)=P(B|A)P(A)
 Conditional Probability:
 Let A and B are events,
 P(A|B)is the conditional probability of A given B,
 P(B|A)is the conditional probability of B given A.
 Equating right and side of both the equations will get:
The above equation (a) is called as Bayes' rule or Bayes' theorem.

This equation is basic of most modern AI systems for probabilistic
inference.
by Ramya p
 P(A|B)is known as posterior, is the Probability of hypothes is A

when occurred an evidence B.
 P(B|A) is called the likelihood, in which hypothesis is true,t hen
calculate the probability of evidence.
 P(A) is called the prior probability, probability of hypothesis before
considering the evidence
 P(B) is called marginal probability, pure probability of an
evidence.
10. Suppose we want to perceive the effect of some unknown cause,

and want to compute that cause, then the Bayes' rule becomes:
what is the probability that a patient has diseases with meningitis

a stiff neck?
Given Data:
A doctor is aware that disease meningitis causes a patient to have a
stiff neck, and it occurs 80%ofthetime.He is also aware of some
more facts, which are given as follows:
o The Known probability that a patient has meningitis disease is
1/30,000.
o The Known probability that a patient has a stiff neck is 2%.
Solution
Let a be the proposition that patient has stiff neck and b be the
proposition that patient has meningitis.
So, calculate the following as:
P(a|b) = 0.8
P(b)=1/30000
P(a)= .02
Hence, assume that1patient out of 750 patients has mening it is

disease with a stiff neck.
11. Considertwoevents:A(itwillraintomorrow)andB(thesunwillshin
etomorrow).
 Use Bayes’ theorem to compute the posterior probability of each event
occurring, given the resulting weather conditions for today: P(A|
sunny)= P(sunny|A)*P(A)/P(sunny)
by Ramya p
P(B|sunny)=P(sunny|B)*P(B)/P(sunny)
where sunny is our evidence (the resulting weather condition for today).
12. What are the Application of Bayes' theorem in Artificial

intelligence?
 It is used to calculate the next step of the robot when the already
executed step is given.
 Bayes' theorem is helpful in weather forecasting.
13. Define Bayesian Network.

 "A Bayesian network is a probabilistic graphical model which
represents a set of variables and their conditional dependencies using a
directed acyclic graph."
 It is also called a Bayes network, belief network, decision network, or
Bayesian model.
 Bayesian Network can be used for building models from data and
experts opinions, and it consists of two parts:
o Directed Acyclic Graph
o Table of conditional probabilities
14. Define Joint probability distribution.

 If variables are x1, x2, x3,....., xn, then the probabilities of a different
combination of x1, x2, x3.. xn, are known as Joint probability
distribution.
 P[x1,x2,x3,,xn],can be written as the following way in terms of the
Joint probability distribution.
=P[x1|x2,x3,.....,xn].p[x2,x3,..............,xn]
=P[x1|x2,x3,.....,xn]P[x2|x3,.....,xn]............P[xn-1|xn]P[xn].
 In general for each variable Xi,
P(Xi|Xi-1,............,X1)=P(Xi|Parents(Xi))
15. Write an algorithm for Constructing Bayesian Network
by Ramya p
16. Define Global semantics and local semantics.

Global Semantics
Local Semantics
17. List the ways to understand the semantics of Bayesian Network

There are two ways to understand the semantics of the Bayesian network,
which is given below:
1. To understand the network as the representation of the Joint
probability distribution.
It is helpful to understand how to construct the network.
2. To understand the network as an encoding of a collection of
conditional independence statements.
It is helpful in designing inference procedure.
18. What are the Applications of Bayesian networks in AI?

1. Spam filtering
2. Bio monitoring
3. Information retrieval
4. Image processing
5. Gene regulatory network
6. Turbo code
7. Document classification
19. Define Bayesian Inference.

 Bayesian Network is to perform inference, which computes the
marginal probability P(V=v) for each node V and each possible
instantiation v.
 Inference can also be done on a Bayesian network when the values of
some nodes are known (as evidence) and wish to compute the
likelihood of values of other nodes.
 There are two types of inference on Bayesian networks: exact and
approximate.
 Exact inference algorithms compute the exact values of each marginal
or posterior probability, while approximate inference algorithms
sacrifice some accuracy of the probabilities to report results quickly.
by Ramya p
20. Define Exact Inference.

 The goal of an exact inference algorithm is to report the exact values for
either the marginal (P(V = v)) or posterior probabilities (P(V = v|e))for
each instantiation v of each node V, possible given some evidence e of
other node values.
21. List the common exact inference algorithms–

 Pearl’s algorithm
 Lauritzen-Spiegelhalter algorithm.
22. Define Pearl’s Algorithm.

 Pearl’s algorithm is a linear-time algorithm that computes the posterior
probabilities of each node given evidence of singly-connected
networks.
 Pearl introduced the notation λ (X =x) for the diagnostic support ofa node
X with value x, which is the probability of evidence below X given that X
= x.
23. Define Lauritzen-Spiegelhalter(LS)Algorithm.

 The Lauritzen-Spiegelhalter (LS) algorithm is an inference algorithm
for Bayesian networks that works on all models.
24. Define Causal Network or Causal Bayesian Network

 A causal network is an acyclic digraph a rising from an evolution of a
substitution system, and representing its history.
 In an evolution of a multi way system, each substitution event is a
vertex in a causal network.
 Two events which are related by causal dependence, meaning one
occurs just before the other, have an edge between the corresponding
vertices in the causal network.
 More precisely, the edge is a directed edge leading from the past event
to the future event.
25. Define Structural Causal Models(SCMs).

 SCMs consist of two parts: a graph, which visualizes causal connections,
and equations, which express the details of the connections. A graph is a
mathematical construction that consists of vertices (nodes)
and edges(links).
 SCMs use a special kind of graph, called a Directed Acyclic Graph
(DAG), for which all edges are directed and no cycles exist.
 DAGs are a common starting place for causal inference.
by Ramya p
26. List the purpose of do-operator in causal networks.

 The do-operator is a mathematical representation of a physical
intervention.
 If the model starts with Z → X → Y, simulate an intervention in X by
deleting all the incoming arrows to X, and manually setting X to some
value x_0.
27. List the rules of Do-Calculus.
8
PREPAREDBY:R.Mohan,G.Selvapriya,S.Shanmuganathan,AP/CSE
by Ramya p
PARTB
1. Explain the concept of uncertainty and acting under uncertainty with suitable
example. Explain in detail about probabilistic reasoning.
UNCERTAINITY&PROBABILISTICREASONING
Uncertainty:
Causes of uncertainty
Probabilistic reasoning:
Need of probabilistic reasoning in AI
Ways to solve problems with uncertain
knowledge
Probability
Conditional probability
Example
Agentsalmostneverhaveaccesstothewholetruthabouttheirenvironment. Agents must,

therefore,act under uncertainty.
Handlinguncertainknowledge
In thissection, welookmoreclosely at the natureofuncertain knowledge. We
will use a simple diagnosis example to illustrate the concepts
involved.Diagnosiswhetherformedicine,automobilerepair,orwhatever-isataskthat
almost always involves uncertainty. Let us try to write rules for dental diagnosisusing
first-order logic, so that we can see how the logical approach breaks down. Consider
the following rule:
The problem is that this rule is wrong. Not all patients with toothaches have
cavities; some of them have gum disease, an abscess, or one of several other
problems:
Unfortunately, in order to make the rule true, we have to add an almost unlimited
list of possible causes. We couldtry turning the rule into a causal rule:
9
PREPAREDBY:R .Mohan,G.Selvapriya,S.Shanmuganathan,AP/CSE
by Ramya p
But this rule is not right either;notall cavities cause pain The only way to fix the rule
is to make it logically exhaustive: to augment the left-hand side with all the
qualificationsrequiredforacavitytocauseatoothache.Eventhen,forthe
purposesofdiagnosis,onemustalsotakeintoaccountthepossibilitythatthe
patientmighthaveatoothacheandacavitythatareunconnected.Tryingtouse first-order
logic to cope with a domain like medical diagnosis thus fails for threemain reasons:
0Laziness: It is too much work to list the complete set of antecedents or

consequentsneededtoensureanexceptionlessruleandtoohardtousesuch rules.
0Theoreticalignorance:Medicalsciencehasnocompletetheoryforthedomain.
0Practicalignorance:Evenifweknowalltherules,wemightbeuncertainabout
aparticularpatientbecausenotallthenecessarytestshavebeenorcanberun.
The connection between toothaches and cavities is just not a logical

consequence in either direction. This is typical of the medical domain, as well
asmostotherjudgmentaldomains:law,business,design,automobilerepair, gardening,
dating, and so on. The agent's knowledge can at best provide only a
degreeofbeliefintherelevantsentences.Ourmaintoolfordealingwithdegreesof belief
will beprobabilitytheory, which assigns to each sentence a numerical degree of
belief between 0 and 1.
Probabilityprovidesawayofsummarizingtheuncertaintythatcomesfromourl
azinessandignorance. We might not know for sure what afflicts a particular
patient,butwebelievethatthereis,say,an80%chance-thatis,aprobabilityof0.8-
thatthepatienthasacavityifheorshehasatoothache.
Thatis,weexpectthatoutofallthesituationsthatareindistinguishable
fromthecurrentsituationasfarastheagent'sknowledgegoes,thepatientwill have a cavity
in 80% of them. This belief could be derived from statistical data-80%
ofthetoothachepatientsseensofarhavehadcavities-orfromsomegeneralrules, or from a
combination of evidence sources.
The80%summarizesthosecasesinwhichallthefactorsneededforacavity to
cause a toothache are present and other cases in which the patient has both
toothache and cavity but the two are unconnected. The missing 20% summarizes
alltheotherpossiblecausesoftoothachethatwearetoolazyorignoranttoconfirm or
deny.
10
by Ramya p
Designforadecision-theoreticagent
Belowalgorithmsketchesthestructureofanagentthatusesdecisiontheory
toselectactions.Theagentisidentical,atanabstractlevel,tothelogicalagent.
Theprimarydifferenceisthatthedecision-theoreticagent'sknowledgeofthe current state
is uncertain; the agent'sbeliefstate is a representation of the probabilities of all
possible actual states of the world. As time passes, the agent accumulates more
evidence and its belief state changes. Given the belief state, the agent can make
probabilistic predictions of action outcomes and hence select the action with highest
expected utility.
Uncertainty:
The knowledge representation, A→B, means if A is true then B is true, but a situation where not sure about whether
So to represent uncertain knowledge, uncertain reasoning or probabilistic reasoning is used.
Causesofuncertainty:
Causesofuncertaintyintherealworld
1. Informationoccurredfromunreliablesources.
2. ExperimentalErrors
3. Equipmentfault
4. Temperaturevariation
5. Climatechange.
Probabilisticreasoning:
 Probabilistic reasoning is a way of knowledge representation, the
concept of probability is applied to indicate the uncertainty in
knowledge.
NeedofprobabilisticreasoninginAI:
o Whenthereareunpredictableoutcomes.
11
by Ramya p
o Whenspecificationsorpossibilitiesofpredicatesbecomestoo large
to handle.
o Whenanunknownerroroccursduringanexperiment.
Waystosolveproblemswithuncertainknowledge:
o Bayes'rule
o BayesianStatistics
Probability:
 Probability can be defined as a chance that an uncertain event
will occur.
 Thevalueofprobabilityalwaysremainsbetween0and1that represent
ideal uncertainties.
o 0≤P(A)≤1, whereP(A)istheprobabilityofaneventA.
o P(A)=0,indicatestotaluncertaintyinaneventA.
o P(A)=1,indicatestotalcertaintyin aneventA.
 Formulatofindtheprobabilityofanuncertainevent
P(¬A) = probabilityofa nothappening event. P(¬A)

+ P(A) = 1.
o Event:Eachpossibleoutcomeofavariableiscalledanevent.
o Samplespace: The collection of all possible events is called
sample space.
o Randomvariables: Randomvariablesareusedtorepresentthe
eventsandobjectsintherealworld.
o Priorprobability: The prior probability of an event is
probabilitycomputedbeforeobservingnewinformation.
o PosteriorProbability:The probability that is calculated after
all evidence or information has taken into account. It is a
combinationofpriorprobabilityandnewinformation.
Conditionalprobability:
 Conditionalprobabilityisaprobabilityofoccurringaneventwhen another
event has already happened.
12
by Ramya p
 Let'ssuppose,tocalculatetheeventAwheneventBhasalready occurred,
"the probability of A under the conditions of B", it is:
Where P(A⋀B)= Joint probability of a and B

P(B)= Marginal probability of B.
 If the probability of A is given and to find the probability of B, then it

is:
Example:
In a class, there are 70% of the students who like English and
40%ofthestudentswholikesEnglishandmathematics,andthenwhatisthe
percent of students those who like English also like mathematics?
Solution:
Let,AisaneventthatastudentlikesMathematics B is
an event that a student likes English.
Hence,57%arethestudentswholikeEnglishalsolikeMathematics
2. ExplainindetailaboutBayesianinferenceandNaiveBayesModelorNaiveBayesThe
oremorBayesRule.
NaiveBayesModelorNaiveBayesTheoremorBayesRule
BayesianInference
BayesTheoremorBayesRule
Example-ApplyingBayes'rule:
ApplicationofBayes'theoreminArtificialintelligence
BayesianInference
 Bayesian inference is a probabilistic approach to machine learning that
provides estimates of the probability of specific events.
 Bayesianinferenceisastatisticalmethodforunderstandingthe
uncertainty inherent in prediction problems.
13
by Ramya p
 BayesianinferencealgorithmcanbeviewedasaMarkovChainMonte
Carloalgorithmthatusespriorprobabilitydistributionstooptimizethe
likelihood function.
 The basis of Bayesian inference is the notion of apriori and a posteriori
probabilities.
o The priori probability is the probability of an event before any
evidence is considered.
o The posteriori probability is the probability of an event after
taking into account all available evidence.
 For example, if we want to know the probability that it will rain
tomorrow, our priori probability would be based on our knowledge of
the weather patterns in our area.
BayesTheoremorBayesRule
 Bayes' theorem can be derived using product rule and conditional
probability of event A with known event B:
 ProductRule:
3. P(A⋀B)=P(A|B)P(B)or
4. P(A⋀B)=P(B|A)P(A)
ConditionalProbability:
LetAandBareevents,
P(A|B)istheconditionalprobabilityofAgivenB,
P(B|A)istheconditionalprobabilityofBgivenA.
Equatingrighthandsideofboththeequationswillget:
The above equation (a) is called asBayes'ruleorBayes'theorem. This

equation is basic of most modern AI systems for
probabilisticinference.
 P(A|B)isknownasposterior,istheProbabilityofhypothesisA when
occurred an evidence B.
 P(B|A) is called thelikelihood, in which hypothesis is true,then
calculate the probability of evidence.
 P(A) is called the priorprobability, probability of hypothesis before
consideringtheevidence
 P(B) is called marginal probability, pure probability of an
evidence.
 Ingeneral,
P(B)=P(A)*P(B|Ai),
14
by Ramya p
 HencetheBayes'rulecanbewrittenas:
WhereA1,A2,A3,........,Anisasetofmutuallyexclusiveand exhaustive
events.
2.3Example1-ApplyingBayes'rule:
Supposewewanttoperceivetheeffectofsomeunknowncause,andw
anttocomputethatcause,thentheBayes'rulebecomes:
whatistheprobabilitythatapatienthasdiseasesmeningitis
withastiffneck?
GivenData:
Adoctorisawarethatdiseasemeningitiscausesapatienttohaveasti
ffneck,anditoccurs80%ofthetime.Heisalsoawareofsomemorefa
cts,whicharegivenasfollows:
TheKnownprobabilitythatapatienthasmeningitisdisease
is1/30,000.
TheKnownprobabilitythatapatienthasastiffneckis2%.
Solution
Letabethepropositionthatpatienthasstiffneckandbbethe
proposition that patient has meningitis.
So,calculatethefollowingas:
P(a|b)=0.8
P(b)=1/30000
P(a)= .02
Hence,assumethat1patientoutof750patientshasmeningitis disease
with a stiff neck.
15
by Ramya p
Example2-ApplyingBayes'rule:
 Considertwoevents:A(itwillraintomorrow)and B(thesunwillshine
tomorrow).
 UseBayes’theoremtocomputetheposteriorprobabilityofeachevent
occurring,giventhe resultingweather conditionsfor today: P(A|
sunny)= P(sunny|A)*P(A)/P(sunny) P(B|sunny)
= P(sunny|B) * P(B) /P(sunny)
wheresunnyisourevidence(theresultingweatherconditionfor today).
 Fromtheseequations,
o if event A is more likely to result in sunny weather than event B,
then the posterior probability of A occurring, given that theresulting
weather condition for today is sunny, will be higher than the
posterior probability of B occurring.
o Conversely, if event Bis more likely to result in sunny weather than
event A, then theposteriorprobabilityof B occurring, given that the
resulting weather condition for today is sunny, will be higher than
the posterior probability of A occurring.
2.4ApplicationofBayes'theoreminArtificialintelligence:
 Itisusedtocalculatethenextstepoftherobotwhenthealready executed
step is given.
 Bayes'theoremishelpfulinweatherforecasting.
NaiveBayesTheorem
The dentistry example illustrates a commonly occurring pattern in which a single
cause directly influences a number of effects, all of which are conditionally
independent, given the cause. The full joint distribution can be written as
Such a probability distribution is called a naive Bayes model—“naive” because it is

often used (as a simplifying assumption)in cases where the “effect” variables are not
strictly independent given the cause variable. (The naive Bayes model is sometimes
called a Bayesian classifier, a somewhat careless usage that has prompted true
Bayesians to call it the idiot Bayes model.) In practice, naive Bayes systems often
work very well, even when the conditional independence assumption is not strictly
true
16
by Ramya p
3. ExplainindetailaboutBayesianNetwork
BayesianNetwork
BayesianNetwork
Joint probabilitydistribution:
ConstructingBayesianNetwork
Example
ThesemanticsofBayesianNetwork
ApplicationsofBayesiannetworksinAI
BayesianNetwork
 "A Bayesian network is a probabilistic graphical model which represents a
setofvariablesandtheirconditionaldependenciesusingadirected acyclic
graph."
 Itisalsocalleda Bayesnetwork,beliefnetwork,decisionnetwork, or Bayesian
model.
 BayesianNetworkcanbeusedforbuildingmodelsfromdataandexperts
opinions,and it consists of two parts:
o DirectedAcyclicGraph
o Tableofconditionalprobabilities
 The generalized form of Bayesian network that represents and solve
decisionproblems under uncertain knowledge is known as an Influence
diagram.
 Itisusedtorepresentconditionaldependencies.
 It can also be used in various tasks including prediction, anomaly
detection, diagnostics, automated insight, reasoning, time series
prediction, and decision making under uncertainty.
 A Bayesian networkgraph ismadeupofnodesandArcs(directedlinks).
Figure2.1–ExampleforBayesianNetwork
 Each node corresponds to the random variables, and a variable can be
continuous or discrete.
17
by Ramya p
 Arc or directed arrows represent the causal relationship or conditional

probabilitiesbetween randomvariables.
 Thesedirectedlinksorarrowsconnectthepairofnodesinthegraph.
 Theselinksrepresentthatonenodedirectlyinfluencetheothernode,and if
there is no directed link that means that nodes are independent with
each other.
Example
In thefigure 2.1,A, B, C, and D are random variablesrepresented by the
nodes of the network graph.
 Considering node B, which is connected with node A by a directed
arrow, then node A is called the parent of Node B.
 NodeCisindependentofnodeA.
 The Bayesian network graph does not contain any cyclic graph. Hence,
it is known as a directed acyclic graph or DAG.
 TheBayesiannetworkhasmainlytwocomponents:
1. CausalComponent
2. Actualnumbers
 EachnodeintheBayesiannetworkhasconditionprobability distribution P(Xi|

Parent(Xi)), which determines the effect of the parent on that node.
 Bayesian network is based on Joint probability distribution and
conditional probability.
Jointprobabilitydistribution:
 If variables are x1, x2, x3,....., xn, then the probabilities of a different
combination of x1, x2, x3.. xn, are known as Joint probability
distribution.
 P[x1,x2,x3,,xn],canbewrittenasthefollowingwayintermsofthe
jointprobability distribution.
=P[x1|x2,x3,.....,xn].p[x2,x3,..............,xn]
=P[x1|x2,x3,.....,xn]P[x2|x3,.....,xn]............P[xn-1|xn]P[xn].
 IngeneralforeachvariableXi,
P(Xi|Xi-1,............,X1)=P(Xi|Parents(Xi))
ConstructingBayesianNetwork
18
by Ramya p
GlobalSemantics
LocalSemantics
MarkovBlanket
 Eachnodeis conditionallyindependentofall othersgivenits
Markovblanket:parents+children+children’sparents
Example:
Harry installed a new burglar alarm at his home to detect burglary.
The alarmreliably responds atdetecting aburglary butalso responds
for minor earthquakes. Harry has two neighbors David and Sophia,
who have taken a responsibility to inform Harry at work when they
hear the alarm. David always calls Harry when he hears the alarm,
but sometimes he got confused with the phone ringing and calls at
that time too.Ontheother hand,Sophia likes to listento high music, so
sometimes she misses to hear the alarm. Here we would like to
computetheprobabilityofBurglaryAlarm.
Problem:
Calculatetheprobabilitythatalarmhassounded,butthereisneitherab
urglary,noranearthquakeoccurred,andDavidandSophiabothcalledt
heHarry.
19
by Ramya p
Solution:
 The Bayesian network for the above problem is given in figure 2.2.
The network structure is showing that burglary and earthquake
istheparentnodeofthealarmanddirectlyaffectingtheprobability of
alarm's going off, but David and Sophia's calls depend on alarm
probability.
Variables:
Burglar,Earthquake,Alarm,JohnCalls,MaryCalls
Networktopologyreflects“causal”knowledge
Aburglarcansetthealarmoff
Anearthquakecansetthealarmoff
ThealarmcancauseMarytocall
The alarmcancauseJohnto call
Figure2.2-TheBayesiannetworkfortheexampleproblem
20
by Ramya p
Alleventsoccurringinthisnetwork:
o Burglary(B)
o Earthquake(E)
o Alarm(A)
o DavidCalls(D)
o Sophiacalls(S)
Writetheeventsofproblemstatementintheformofprobability:
P[D,S,A,B,E],
Rewritetheprobabilitystatementusingjointprobabilitydistribution:
Let's take the observed probability for the Burglary and earthquake
component:
 P(B=True)=0.002,whichistheprobabilityofburglary.
 P(B=False)=0.998,whichistheprobabilityofnoburglary.
 P(E=True)=0.001,whichistheprobabilityofaminorearthquake
 P(E=False)=0.999,Whichistheprobabilitythatanearthquakenot occurred.
ConditionalprobabilitytableforAlarmA:
TheConditionalprobabilityofAlarmAdependsonBurglarand earthquake:
B E P(A=True) P(A=False)
True True 0.94 0.06
True False 0.95 0.04
False True 0.31 0.69
False False 0.001 0.999
ConditionalprobabilitytableforDavidCalls:
The Conditionalprobability ofDavid thathewillcall dependsonthe
probability of Alarm.
21
by Ramya p
A P(D=True) P(D=False)
True 0.91 0.09
False 0.05 0.95
ConditionalprobabilitytableforSophiaCalls:
TheConditionalprobabilityofSophiathatshecallsisdependingon its Parent
Node "Alarm."
A P(S=True) P(S=False)
True 0.75 0.25
False 0.02 0.98
Fromtheformulaof joint distribution,theproblemstatement in theformof

probability distribution:
P(S,D,A,¬B,¬E)=P(S|A)*P(D|A)*P(A|¬B^¬E)*P(¬B)*P(¬E).
=0.75*0.91*0.001*0.998*0.999
=0.00068045.
Hence, a Bayesian network can answer any query about the domain
by using Joint distribution.
ThesemanticsofBayesianNetwork:
TherearetwowaystounderstandthesemanticsoftheBayesiannetwork, which is
given below:
1. TounderstandthenetworkastherepresentationoftheJointproba
bilitydistribution.
Itishelpfultounderstandhowtoconstructthenetwork.
2. Tounderstandthenetworkasanencodingofacollectionofcondit
ionalindependencestatements.
Itishelpfulindesigninginference procedure.
ApplicationsofBayesiannetworksinAI
Bayesiannetworksfindapplicationsinavarietyoftaskssuchas:
1. Spamfiltering:
a. A spam filter is a program that helps in detecting unsolicited and
spammails.Bayesianspamfilterscheckwhetheramailisspam or not.
22
by Ramya p
2. Biomonitoring:
a. Thisinvolvestheuseofindicatorstoquantifytheconcentrationof
chemicals in the human body.
3. Informationretrieval:
a. Bayesian networks assist in information retrieval for research,which
is a constant process of extracting information fromdatabases.
4. Imageprocessing:
a. A form of signal processing, image processing uses mathematical
operations to convert images into digital format.
5. Generegulatorynetwork:
a. A Bayesian network is an algorithm that can be applied to gene
regulatory networks in order to make predictions about the effectsof
genetic variations on cellular phenotypes.
b. Generegulatorynetworksareasetofmathematicalequationsthat
describetheinteractionsbetweengenes,proteins,andmetabolites.
c. They are used to study how genetic variations affect the
development of a cell or organism.
6. Turbocode:
a. Turbo codesare a typeof error correctioncode capableof achieving
very high data rates and long distances between error correcting
nodes in a communications system.
b. They have been used in satellites, space probes, deep-space
missions, military communications systems, and civilian wireless
communication systems, including WiFi and 4G LTE cellular
telephone systems.
7. Documentclassification:
a. The main issue is to assign a document multiple classes. The task
can be achieved manually and algorithmically. Since manual effort
takes too much time, algorithmic documentation is done tocomplete
it quickly and effectively.
4. ExplainindetailaboutBayesianInferenceanditstypeExactInferencewithsuitabl
eexample.
ExactinferenceinBayesiannetworks
The basic task for any probabilistic inference system is to compute the
posterior probability distribution for a set of query variables, given some observed
event-thatis,someassignmentofvaluestoasetofevidencevariables.Wewilluse
23
by Ramya p
thenotationXdenotesthequeryvariable;EdenotesthesetofevidencevariablesEl, . . . ,Em,and
e is a particular observed event;Y will denote the nonevidencevariablesYl,...,(some-
timescalledthehiddenvariables).Thus,thecompleteset of variables X ={X}U E U Y. A
typical query asks for the posterior probability distribution P(X|e)4
In the burglary network, we might observe the event in whichJohnCalls =

true andMaryCalls =true. We could then ask for, say, the probability that a
burglary has occurred:
Inferencebyenumeration
Conditional probability can be computed by summing terms from the full joint
distribution. More specifically, a query P(X|e) can be answered using Equation,
whichwe repeathereforconvenience:
Now, a Bayesian network gives a complete representation of the full joint

distribution. More specifically, Equation shows that the termsP(x, e, y) in the joint
distributioncanbewrittenasproductsofconditionalprobabilitiesfromthe network.
Therefore,aquerycanbeansweredusingaBayesiannetworkbycomputingsumsof
productsofconditionalprobabibitiesfromthenetwork.InFigure an algorithm,
ENUMERATE-JOINT-ASK, was given for inference by enumeration
fromthefulljointdistribution.Thealgorithm takesasinputafulljointdistribution
Pandlooksupvaluestherein.Itisasimplemattertomodifythealgorithmsothat it takes as
input a Bayesian network bn and "looks up" joint entries by multiplyingthe
corresponding CPT entries from bn.
Consider the query P(Burglary1 JohnCalls = true,&Jury Calls = true). The

hidden variables for this query are Earthquake and Alarm. From Equation (13.6),
using initialletters forthe variables inorder to shortenthe expressions, we have
The semantics of Bayesian networks then gives us an expression in terms of

CPT entries. For simplicity, we will do this just for Burglizry = true:
24
by Ramya p
To compute this expression, we have to add four terms, each computed by

multiplyingfivenumbers.Intheworstcase,wherewehavetosumoutalmostall
thevariables,thecomplexityofthealgorithmforanetworkwithnBoolean variables
isO(n2n). An improvement can be obtained from the following simple
observations:theP(b)termisaconstantandcanbemovedoutsidethe summaltions over a
and e, and the 13(e) term can be moved outside the summation over a. Hence, we
have
This expression can be evaluated by looping through the variables in order,

multiplying CPT entries as we go. For each summation, we also needto loop overthe
variable's possible values. The structure of this computation is shown in Figure.
Using the numbers from Figure, we obtainP(b|j , m) =a x 0.00059224.'The
correspondingc omputation for ~b yields a x 0.0014919; hence
This expression can be evaluated by looping through the variables in order,

multiplying CPT entries as we go. For each summation, we also need to loop overthe
variable's possible values.
ThestructureofthiscomputationisshowninaboveFigure.Usingthenumbers from Figure,

we obtainP(b|j , m) = a x 0.00059224.'The co~respondingc omputation for ~b yields
a x 0.0014919; hence
25
by Ramya p
Thatis,thechanceofaburglary,givencallsfrombothneighbors,isabout28%.
TheevaluationprocessfortheexpressioninEquationisshownasanexpression tree in
Figure.
Thevariableeliminationalgorithm
The enumeration algorithm can be improved substantially by eliminating
repeated calculations of the kind illustrated in Figure. The idea is simple: do the
calculation once and save the results for later use. This is a form of dynamic
programming. There are several versions of this approach; we present the variable
elimination algorithm, which is the simplest. Variable elimination works by
evaluatingexpressionssuch as Equationinright-to-left order (thatis,bottom-upin
Figure).Intermediateresultsarestored,andsummationsovereachvariableare done only
for those portions of the expression that depend on the variable. Let us illustrate this
process for the burglary network. We evaluate the expression
Thecomplexityofexactinference
We have argued that variable elimination is more efficient than enumeration

because it avoids repeated computations (as well as dropping irrelevant variables).
Thetimeandspacerequirementsof variable eliminationare dominatedby thesize
ofthelargestfactorconstructedduringtheoperationofthealgorithm.Thisinturn is
determined by the order of elimination of variables and by the structure of the
network.
TheburglarynetworkofFigurebelongstothefamilyofnetworksinwhich
thereisatmostoneundirectedpathbetweenanytwonodesinthenetwork.These are
calledsinglyconnected networks orpolytrees, and they have a particularly nice
property:
Thetimeandspacecomplexityofexactinferenceinpolytreesislinearinthesizeofthenetw
ork.I-Iere,thesizeisdefinedasthenumberofCPTentries;if
26
by Ramya p
the number of parents of each node is bounded by a constant, then the complexity
will also be linear in the number of nodes. These results hold for any ordering
consistent with the topological ordering of the network .
Formultiplyconnected networks, such as that of Figure, variable

elimination can have exponential time and space complexity in the worst case, even
when the number of parents per node is bounded. This is not surprising when one
considersthat,becauseitincludesinferenceinpropositionallogicasaspecialcase,infer
enceinBayesiannetworksis1W-hard. In fact, it can be shown that the problem is
as hard as that of computing thenumber of satisfying assignments for a propositional
logic formula. This means that it is #P-hard ("number-P hard")-that is, strictly harder
than NP-complete problems.
There is a close connection between the complexity of Bayesian network

inference and the complexity of constraint satisfaction problems (CSPs), thedifficulty
of solving a discrete CSP is related to how"tree-lilce" its constraint graphis Measures
such ashypertreewidth, which bound the complexity of solving a CSP, can also
be applied directly to Bayesian networks. Moreover, the variable elimination
algorithm can be generalized to solve CSPs as well as Bayesiannetworks.
5. ExplainCausalNetworkorCausalBayesianNetworkinMachine
CausalNetworkorCausalBayesianNetwork
 Acausalnetworkisanacyclicdigrapharisingfromanevolutionof
asubstitutionsystem,andrepresentingitshistory.
 Inanevolution ofa multiwaysystem,eachsubstitutioneventisavertex in a
causal network.
 Two events which are related by causal dependence, meaning one occurs
just before the other, have an edge between the corresponding vertices in
the causal network.
 More precisely, the edge is a directed edge leading from the past event to
the future event.
 ReferFigure2.3foranexamplecausalnetwork.
 A CBN is a graph formed by nodes representing random variables,
connectedbylinksdenotingcausalinfluence.
27
by Ramya p
Figure2.3–CausalNetworkExample
 Somecausalnetworksareindependentofthechoiceofevolution,and these
are called causally invariant.
 StructuralCausalModels(SCMs).
 SCMsconsistoftwoparts:agraph,whichvisualizescausal
connections,andequations,whichexpressthedetailsofthe
connections.agraphisamathematicalconstructionthatconsistsofv
ertices(nodes)andedges(links).
 SCMs use a special kind of graph, called
aDirectedAcyclicGraph(DAG), for which all edges are directed and
no cycles exist.
 DAGsareacommonstartingplaceforcausalinference.
 Bayesianandcausalnetworksarecompletelyidentical.However, the
difference lies in their interpretations.
Fire->Smoke
 Anetworkwith2nodes(fireiconandsmokeicon)and1edge(arrow pointing
from fire to smoke).
 ThisnetworkcanbebothaBayesianorcausalnetwork.
 Thekeydistinction, however,iswheninterpreting thisnetwork.
 For a Bayesian network, we view the nodesasvariables and the
arrowasaconditionalprobability, namely the probability of smoke given
informationabout fire.
 Wheninterpretingthis as a causal network, westillview
nodesasvariables,however,thearrowindicatesacausalconnection.
28
by Ramya p
 In this case, both interpretations are valid. However, if we were to flip the
edge direction, the causal network interpretation would be invalid, since
smoke does not cause fire.
ImplementingCausalInference
1.Thedo-operator
 Thedo-operatoris
amathematicalrepresentationofaphysicalintervention.
 If the model starts with Z → X → Y, simulate an intervention in X by
deleting all the incoming arrows to X, and manually setting X to some
value x_0. Refer Figure 2.4 denotes the example of do-operator.
Figure2.4–do-operatorExample
P(Y|X) is the conditional probability that is,

theprobabilityofYgivenanobservationofX.While, P(Y|do(X)) isthe
probabilityofYgivenaninterventioninX.
29
by Ramya p
2:Confounding
Asimpleexampleofconfoundingisshowninthefigure 2.5below.
Figure2.5–ConfoundingExample
 In this example, age is a confounder of education and wealth. In

other words, if trying to evaluate the impact of education on wealth
one would need to adjust for age.
 Adjustingfor (orconditioningon) age justmeansthatwhen
lookingatage,education,andwealthdata,onewouldcompare data points
within age groups, notbetween age groups.
 Confounding isanything that leads to P(Y|X) being different than
P(Y|do(X)).
3:EstimatingCausalEffects
 Treatment effect = (Outcome under E) minus (Outcome under C),
that is the difference between the outcome a child would receive if
assigned to treatment E and the outcome that same child would
receive if assigned to treatment C. These are called potential
outcomes.
6. ExplainapproximateinferenceinBayesiannetwork(BN)
Given the intractability of exact inference in large networks, we will now

consider approximate inference methods. This section describes randomizedsampling
algorithms, also called Monte Carlo algorithms, that provide approximate
answerswhose accuracy dependson thenumberofsamplesgenerated.
They work by generating random events based on the probabilities in the

Bayesnetandcountingupthedifferentanswersfoundinthoserandomevents. With enough
samples, we can get arbitrarily close to recovering the true probability distribution—
provided the Bayes net has no deterministic conditional distributions
30
by Ramya p
Directsamplingmethods
The primitive element in any sampling algorithm is the generation of samples
fromaknownprobabilitydistribution.Forexample,anunbiasedcoincanbe thought of as a
random variableCoin with values (heads,tails) and a prior distribution P(Coin) =
(0.5,0.5). Sampling from this distribution is exactly like
flippingthecoin:withprobability0.5it will returnheads,andwithprobability0.5 it will
return tails.
Given a source of random numbersruniformly distributed in the range [0,1],
itisasimplemattertosampleanydistributiononasinglevariable,whether discrete or
continuous. This is done by constructing the cumulative distribution for the variable
and returning the first value whose cumulative probability exceedsr
We begin with a random sampling process for a Bayes net that has no evidence
associated with it. The idea is to sample each variable in turn, in topological order.
The probability distribution from which the value is sampled is conditioned on the
values already assigned to the variable’s parents. (Because we sample in topological
order,theparentsareguaranteedtohavevaluesalready.)Thisalgorithmisshown in Figure.
Applying it to the network with the orderingCloudy,Sprinkler,Rain, WetGrass,
we might produce a random event as follows:
RejectionsamplinginBayesiannetworks
Rejectionsampling isa general methodforproducing samplesfroma hard- to-
sampledistributiongivenaneasy-to-sampledistribution.Initssimplestform,it can be
used to compute conditional probabilities thatis, to determineP(X|e).The
REJECTION-SAMPLINGalgorithmisshowninFigure.First,itgeneratessamples
from the prior distribution specified by the network. Then, it rejects all those that
donotmatchtheevidence.Finally,theestimateˆP(X=x|e)isobtainedbycounting how
often X=xoccursin the remainingsamples.
31
by Ramya p
Let ˆP(X|e)be the estimated distribution that the algorithm returns; this
distribution is computed by normalizing NPS(X,e), the vector of sample counts for
each value of X where the sample agrees with the evidence e:
InferencebyMarkovchainsimulation
In this section, we describe theMarkovchainMonteCarlo (MCMC)
algorithm for inference in Bayesian networks. We will first describe what the
algorithmdoes,thenwewillexplainwhyitworksandwhyithassucha complicated name.
TheMCMCalgorithm
MCMC generates each event by making a random change to the preceding

event. It is therefore helpful to think of the network as being in a
particularcurrentstate specifying a value for every variable. The next state is
generated by randomly
samplingavalueforoneofthenonevidencevariablesXi,conditionedonthecurrentvalue
softhevariablesintheMarkovblanketofXi. MCMC therefore wanders randomly
around the state space-the space of possible complete assignments-flipping one
variable at a time, but keeping the evidence variables fixed.
ConsiderthequeryP(Rain1Sprinkler=true,WetGrass=true)appliedtothe
network in Figure. The evidence variablesSprinkler andWetGrass are fixed to their
observedvaluesandthehiddenvariablesCloudyandRainareinitializedrandomly- let us say
totrue andfalse respectively. Thus, the initial state is[true,true,false,true]. Now the
following steps are executed repeatedly:
32
by Ramya p
1. Cloudyissampled,giventhecurrentvaluesofitsMarkovblanketvariables:in this case,

we sample fromP(Cloudy1Sprinkler =true,Rain=false). Suppose the result is
Cloudy=false. Then the new current state is [false,true,false,true].
2. Rain is sampled, given the current values of its Markov blanket variables: in this
case, we sample from P(Rain1Cloudy=false, Sprinkler=true, Wet Grass=true).
Suppose this yields Rain = true. The new current state is [false, true, true,
true].Each state visited during this process is a sample that contributes to the
estimate for the query variable Rain. If the process visits 20 states where Rain is
true and 60 states where Rain is false, then the answer to the query is
NORMALIZE
33
49
Downloaded by Ramya p (ramyafeb35@gmail.com)

Aiml Unit 2

Uploaded by

Copyright:

Available Formats

Aiml Unit 2

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Aiml Unit 2

Uploaded by

Copyright:

Available Formats

CS3491– ARTIFICIALINTELLIGENCEANDMACHINELEARNING UNIT2

2. Define Probabilistic reasoning. Mention the need of probabilistic

3. List the Ways to solve problems with uncertain knowledge.

4. Define Probability and the probability of occurrence.

P(¬A) = probability of a not happening event. (¬A) + P(A) = 1.

5. Define the terms event, sample space, random variables, prior

6. Define Conditional probability.

Where P(A⋀B)= Joint probability of a and B

8. Define Bayesian Inference.

9. List Bayes Theorem or Bayes Rule

The above equation (a) is called as Bayes' rule or Bayes' theorem.

 P(A|B)is known as posterior, is the Probability of hypothes is A

10. Suppose we want to perceive the effect of some unknown cause,

what is the probability that a patient has diseases with meningitis

Hence, assume that1patient out of 750 patients has mening it is

12. What are the Application of Bayes' theorem in Artificial

13. Define Bayesian Network.

14. Define Joint probability distribution.

15. Write an algorithm for Constructing Bayesian Network

16. Define Global semantics and local semantics.

17. List the ways to understand the semantics of Bayesian Network

18. What are the Applications of Bayesian networks in AI?

19. Define Bayesian Inference.

20. Define Exact Inference.

21. List the common exact inference algorithms–

22. Define Pearl’s Algorithm.

23. Define Lauritzen-Spiegelhalter(LS)Algorithm.

24. Define Causal Network or Causal Bayesian Network

25. Define Structural Causal Models(SCMs).

26. List the purpose of do-operator in causal networks.

27. List the rules of Do-Calculus.

Agentsalmostneverhaveaccesstothewholetruthabouttheirenvironment. Agents must,

0Laziness: It is too much work to list the complete set of antecedents or

The connection between toothaches and cavities is just not a logical

P(¬A) = probabilityofa nothappening event. P(¬A)

Where P(A⋀B)= Joint probability of a and B

 If the probability of A is given and to find the probability of B, then it

The above equation (a) is called asBayes'ruleorBayes'theorem. This

Such a probability distribution is called a naive Bayes model—“naive” because it is

 Arc or directed arrows represent the causal relationship or conditional

 EachnodeintheBayesiannetworkhasconditionprobability distribution P(Xi|

True True 0.94 0.06

True False 0.95 0.04

False True 0.31 0.69

False False 0.001 0.999

True 0.91 0.09

False 0.05 0.95

True 0.75 0.25

False 0.02 0.98

Fromtheformulaof joint distribution,theproblemstatement in theformof

In the burglary network, we might observe the event in whichJohnCalls =

Now, a Bayesian network gives a complete representation of the full joint

Consider the query P(Burglary1 JohnCalls = true,&Jury Calls = true). The

The semantics of Bayesian networks then gives us an expression in terms of

To compute this expression, we have to add four terms, each computed by

This expression can be evaluated by looping through the variables in order,

This expression can be evaluated by looping through the variables in order,

ThestructureofthiscomputationisshowninaboveFigure.Usingthenumbers from Figure,

We have argued that variable elimination is more efficient than enumeration