Understanding Bayesian Networks Basics
Translated by
ScribdTranslationsUnderstanding Bayesian Networks Basics
Translated by
ScribdTranslationsBAYESIAN NETWORKS
NELSON CAMPAGNARO JR
JOINVILLE, APRIL 2011
Summary
Introduction
Bayesian Networks - Concepts
Probability
Structure of a Bayesian Network
Knowledge Representation
Inference using Bayesian Networks
Semantics
Applications
Building a Bayesian Network
Conclusion
Exercises
References
Introduction
Bayesian Networks were developed
early 1980s to make the task easier of
prediction and "abduction" in systems of
Artificial Intelligence (AI) (Pearl, 2000). In
summary, Bayesian Networks (BN) also
known as opinion networks, networks
causal, dependency graphs
probabilistic, they are graphical models for
reasoning (conclusions) based on uncertainty,
where the nodes represent the variables
(discrete or continuous), and the arcs
they represent the direct connection between them.
Introduction
Bayesian networks are models of
representation of knowledge that
they work with uncertain knowledge and
incomplete through the Theory of
Bayesian Probability, published by
mathematician Thomas Bayes in 1763.
It has been becoming the standard methodology.
for the construction of trust systems
in probabilistic knowledge and has been
applied in a variety of activities of
real world.
Bayesian Networks
They are diagrams that organize the
knowledge in a given area through
a mapping between causes and effects.
Network-based systems
Bayesian are capable of generating
automatically predictions or decisions
even in the situation of non-existence of
some pieces of information.
Bayesian Networks
There are two main approaches that
can be used within the context
two systems that act rationally:
Logical Reasoning
Probabilistic Reasoning
Logical and Probabilistic Reasoning
Reflect on prior knowledge
respect for the problem and, on this basis
knowledge draws its conclusions.
Bayesian networks offer a
approach to reasoning
probabilistic that encompasses theory of
graphs, for the establishment of
relations between sentences and also, theory
of probabilities.
Probability
Conditional probability deals with
probability of an event A occurring,
having occurred an eventB, both of
sample spaces, that is, it is calculated
sobre o eventoBe não em função o
sample spaces.
The probability of occurrence of a
event Regarding an event
occurrenceBé expresses as:
P(A/B)
Probability
Provides a way to describe and manipulate
uncertain or incomplete knowledge.
Associate the sentences with a degree of belief
numeric between 0 and 1.
Every statement is either true or false.
Bayes' theorem
Thomas Bayes
Probability
Formula:
Degree of Probability
Conditional: calculated according to the
available evidence. Given two events
A and B, the conditional probability of A
Given that B occurred, it is denoted by P(A/B)
is defined by:
P(A B)
P(A|B) , P(B) 0.
P(B)
P(cavity|toothache) = 0.5
Degree of Probability
Independence of Events: Two events A
A and B are independent if the information of
the occurrence (or non-occurrence) of B does not alter the
probability of occurrence of A, that is:
P(A | B) P(A)
Ex:A= dor de ouvido , B= úlcera, ou seja
ulcer does not cause earache.
Probability Table
The filling of the tables of
conditional probabilities are many
simple times (as long as the relationship between
the parents and the child node should not be arbitrary.
Generally, the relationships between us parents and
we children fall into categories of
canonical distributions (which obey to
a standard), only being necessary to
identify what the pattern is and introduce it
some parameters.
Probability Table
Each line in a Tables of
conditional probability contains the
conditional probability of each value
from it to a case of conditioning.
a case of conditioning is just
a possible combination of values for
the upper knots.
A node without parents has only one line.
Probability Table
Example:
Variáveis:Arrombamento,Terremoto,
Alarm, João connects, Maria connects.
Probability Table
Robberies and earthquakes directly affect
the probability of the alarm going off. But the
the fact that João and Maria only call
it depends on the alarm;
In this way, the network represents our
suppositions that they do not notice
any thefts directly, they do not notice
the earthquakes and do not verify before
to connect.
Structure of a Bayesian network
Each random variable (RV) is
represented by a node of the network
Each node (VA) receives connections from the nodes
who have direct influence (your parents) over
it. (Easy task for the expert)
Each node has a table of
Conditional Probabilities that
quantifies the influence of your parents on
he. (Difficult for the specialist)
The graph is acyclic
Structure of a Bayesian network
Set of discrete variables U = {A1, A2,
...,An}.
Set of directed edges between
variables.
Directed cycles cannot exist
Each variable has a finite set of
mutually exclusive states.
Structure of a Bayesian network
Knowledge representation
for reasoning with uncertainty
Represents 3 types of knowledge
of the domain:
Independence relationships between random variables
(graphically);
Prior probabilities of some variables;
Conditional probabilities between variables
dependents.
Representation of knowledge
for reasoning with uncertainty
It allows efficient calculation:
Posterior probabilities of any
random variable (inference), using for
this is a recursive definition of the theorem
of Bayes.
Knowledge representation
for reasoning with uncertainty
Represented knowledge:
It can be learned from examples,
reusing part of the mechanisms of
reasoning.
Reasoning with Uncertainty
The chance of Flamengo winning the next one
the game is 78%.
The probability of it raining tomorrow is
90%.
The vast majority of Brazilians like to
football.
José thinks that the horse Azulão is going to win.
the match, but Pedro thinks otherwise.
Inference using Bayesian Networks
The joint distribution can be used to
answer any question about the
domain.
Types:
Causal
Diagnosis
Intercausal
Semantics
Two semantics:
Numerical (global)
Topological(local)
Semântica Local
Ex.Dé independent of AeB
Application
Diagnosis of heart diseases: The
Probabilistic Network Technology
(Bayesian Networks) is ideal for the
treatment of uncertainty, very common in
medical area and, furthermore, shapes the
domain expert knowledge
in an intuitive way.
Building a Bayesian Network
Choose an order for the variables
random variables X1,…,Xn.
For i = 1 to n, add Xi to the network, select
country for X1, ..., Xi-1 such that P(Xi |
Country(Xi) = P(Xi | X1, ... Xi-1).
The correct order in which the nodes should be
added consists of adding first
as 'root causes', then the variables that
they influence and so on, until
we reach the leaves.
Building a Bayesian Network
Application Example
The objective here is to extract knowledge.
automatically from a base
of hypothetical data containing a number
candidate sequence and three more
variáveis: Aprovado, Cursinho e IBL
(Broadband Internet), containing two
possible attributes each (Yes and No).
Building a Bayesian Network
APPROVED IBL little course
Yes Yes Yes
Yes No Yes
Yes Yes No
No No No
Yes No Yes
Yes Yes Yes
Yes No Yes
No No No
No Yes No
Yes Yes Yes
No No No
Yes Yes No
Yes No Yes
Yes Yes Yes
No No Yes
Building a Bayesian Network
What is the probability of a candidate
to be approved given that you have internet of
broadband at home?
What is the probability of a candidate being
Approved since you took a preparatory course?
Conclusion
The greatest advantage of reasoning
probabilistic in relation to reasoning
it is logical to allow the agent to arrive at
rational decisions even when there are none
sufficient information to prove that
any of the given actions will work.
Exercises
01)
Mobville, the juggling thief, often drops the balls he is balancing (or
loading) when its battery is low. In previous tests, it was determined that
the probability that he drops a ball when the battery is low is 0.9. Therefore
on the other hand, when the battery is not low, the probability of him dropping a ball
It is only 0.01. The battery was charged a little while ago, and in our best guess,
Given your feats with the balls in the air, having a low battery is ten to one. A robbery.
observer, with a not very reliable vision system, warns that Mobville has knocked down a
ball. The reliability of the observer is given by the following probabilities:
1.p(observer says that Mobville brings down|Mobville brings down) = 0.9
2.p(observer says that Mobville falls|Mobville falls) = 0.2
Draw a Bayes network, and calculate the probability that the battery is low given the
observer report (knocked over a ball).
Exercises
02)
An admissions committee for a program
of the master's degree is trying to determine the
probability that a candidate is admitted
be truly qualified. The odds
relevant are given by the Bayes network
shown in figure 2.
Exercises
Exercises
Determine P(A|D) (the probability of a
qualified candidate (A), given that this
have been accepted into the program of
master's degree (D).
References
[Link]
Alexandre Serra Barreto
[Link] for
Célia Flores and Charles Hoher
[Link] André Hideaki,
Rodolfo Sharovysk, Fábio Gagliardi
[Link]
Felipe Nunes Leonel